Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Simple NUMA memory policy for the Linux kernel. |
| 3 | * |
| 4 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 5 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6 | * Subject to the GNU Public License, version 2. |
| 7 | * |
| 8 | * NUMA policy allows the user to give hints in which node(s) memory should |
| 9 | * be allocated. |
| 10 | * |
| 11 | * Support four policies per VMA and per process: |
| 12 | * |
| 13 | * The VMA policy has priority over the process policy for a page fault. |
| 14 | * |
| 15 | * interleave Allocate memory interleaved over a set of nodes, |
| 16 | * with normal fallback if it fails. |
| 17 | * For VMA based allocations this interleaves based on the |
| 18 | * offset into the backing object or offset into the mapping |
| 19 | * for anonymous memory. For process policy an process counter |
| 20 | * is used. |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 21 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | * bind Only allocate memory on a specific set of nodes, |
| 23 | * no fallback. |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 24 | * FIXME: memory is allocated starting with the first node |
| 25 | * to the last. It would be better if bind would truly restrict |
| 26 | * the allocation to memory nodes instead |
| 27 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | * preferred Try a specific node first before normal fallback. |
David Rientjes | 00ef2d2 | 2013-02-22 16:35:36 -0800 | [diff] [blame] | 29 | * As a special case NUMA_NO_NODE here means do the allocation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | * on the local CPU. This is normally identical to default, |
| 31 | * but useful to set in a VMA when you have a non default |
| 32 | * process policy. |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 33 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | * default Allocate on the local node first, or when on a VMA |
| 35 | * use the process policy. This is what Linux always did |
| 36 | * in a NUMA aware kernel and still does by, ahem, default. |
| 37 | * |
| 38 | * The process policy is applied for most non interrupt memory allocations |
| 39 | * in that process' context. Interrupts ignore the policies and always |
| 40 | * try to allocate on the local CPU. The VMA policy is only applied for memory |
| 41 | * allocations for a VMA in the VM. |
| 42 | * |
| 43 | * Currently there are a few corner cases in swapping where the policy |
| 44 | * is not applied, but the majority should be handled. When process policy |
| 45 | * is used it is not remembered over swap outs/swap ins. |
| 46 | * |
| 47 | * Only the highest zone in the zone hierarchy gets policied. Allocations |
| 48 | * requesting a lower zone just use default policy. This implies that |
| 49 | * on systems with highmem kernel lowmem allocation don't get policied. |
| 50 | * Same with GFP_DMA allocations. |
| 51 | * |
| 52 | * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between |
| 53 | * all users and remembered even when nobody has memory mapped. |
| 54 | */ |
| 55 | |
| 56 | /* Notebook: |
| 57 | fix mmap readahead to honour policy and enable policy for any page cache |
| 58 | object |
| 59 | statistics for bigpages |
| 60 | global policy for page cache? currently it uses process policy. Requires |
| 61 | first item above. |
| 62 | handle mremap for shared memory (currently ignored for the policy) |
| 63 | grows down? |
| 64 | make bind policy root only? It can trigger oom much faster and the |
| 65 | kernel is not always grateful with that. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | */ |
| 67 | |
| 68 | #include <linux/mempolicy.h> |
| 69 | #include <linux/mm.h> |
| 70 | #include <linux/highmem.h> |
| 71 | #include <linux/hugetlb.h> |
| 72 | #include <linux/kernel.h> |
| 73 | #include <linux/sched.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | #include <linux/nodemask.h> |
| 75 | #include <linux/cpuset.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | #include <linux/slab.h> |
| 77 | #include <linux/string.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 78 | #include <linux/export.h> |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 79 | #include <linux/nsproxy.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | #include <linux/interrupt.h> |
| 81 | #include <linux/init.h> |
| 82 | #include <linux/compat.h> |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 83 | #include <linux/swap.h> |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 84 | #include <linux/seq_file.h> |
| 85 | #include <linux/proc_fs.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 86 | #include <linux/migrate.h> |
Hugh Dickins | 62b61f6 | 2009-12-14 17:59:33 -0800 | [diff] [blame] | 87 | #include <linux/ksm.h> |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 88 | #include <linux/rmap.h> |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 89 | #include <linux/security.h> |
Adrian Bunk | dbcb0f1 | 2007-10-16 01:26:26 -0700 | [diff] [blame] | 90 | #include <linux/syscalls.h> |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 91 | #include <linux/ctype.h> |
KOSAKI Motohiro | 6d9c285 | 2009-12-14 17:58:11 -0800 | [diff] [blame] | 92 | #include <linux/mm_inline.h> |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 93 | #include <linux/mmu_notifier.h> |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 94 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | #include <asm/tlbflush.h> |
| 96 | #include <asm/uaccess.h> |
Michal Hocko | 778d3b0 | 2011-07-26 16:08:30 -0700 | [diff] [blame] | 97 | #include <linux/random.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 99 | #include "internal.h" |
| 100 | |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 101 | /* Internal flags */ |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 102 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 103 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 104 | |
Pekka Enberg | fcc234f | 2006-03-22 00:08:13 -0800 | [diff] [blame] | 105 | static struct kmem_cache *policy_cache; |
| 106 | static struct kmem_cache *sn_cache; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | /* Highest zone. An specific allocation for a zone below that is not |
| 109 | policied. */ |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 110 | enum zone_type policy_zone = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 112 | /* |
| 113 | * run-time system-wide default policy => local allocation |
| 114 | */ |
H Hartley Sweeten | e754d79 | 2011-10-31 17:09:23 -0700 | [diff] [blame] | 115 | static struct mempolicy default_policy = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 117 | .mode = MPOL_PREFERRED, |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 118 | .flags = MPOL_F_LOCAL, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | }; |
| 120 | |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 121 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
| 122 | |
| 123 | static struct mempolicy *get_task_policy(struct task_struct *p) |
| 124 | { |
| 125 | struct mempolicy *pol = p->mempolicy; |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 126 | |
| 127 | if (!pol) { |
Jianguo Wu | 1da6f0e | 2013-09-11 14:21:25 -0700 | [diff] [blame] | 128 | int node = numa_node_id(); |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 129 | |
Jianguo Wu | 1da6f0e | 2013-09-11 14:21:25 -0700 | [diff] [blame] | 130 | if (node != NUMA_NO_NODE) { |
| 131 | pol = &preferred_node_policy[node]; |
| 132 | /* |
| 133 | * preferred_node_policy is not initialised early in |
| 134 | * boot |
| 135 | */ |
| 136 | if (!pol->mode) |
| 137 | pol = NULL; |
| 138 | } |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 139 | } |
| 140 | |
| 141 | return pol; |
| 142 | } |
| 143 | |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 144 | static const struct mempolicy_operations { |
| 145 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 146 | /* |
| 147 | * If read-side task has no lock to protect task->mempolicy, write-side |
| 148 | * task will rebind the task->mempolicy by two step. The first step is |
| 149 | * setting all the newly nodes, and the second step is cleaning all the |
| 150 | * disallowed nodes. In this way, we can avoid finding no node to alloc |
| 151 | * page. |
| 152 | * If we have a lock to protect task->mempolicy in read-side, we do |
| 153 | * rebind directly. |
| 154 | * |
| 155 | * step: |
| 156 | * MPOL_REBIND_ONCE - do rebind work at once |
| 157 | * MPOL_REBIND_STEP1 - set all the newly nodes |
| 158 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes |
| 159 | */ |
| 160 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes, |
| 161 | enum mpol_rebind_step step); |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 162 | } mpol_ops[MPOL_MAX]; |
| 163 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 164 | /* Check that the nodemask contains at least one populated zone */ |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 165 | static int is_valid_nodemask(const nodemask_t *nodemask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | { |
Lai Jiangshan | d3eb157 | 2013-02-22 16:33:22 -0800 | [diff] [blame] | 167 | return nodes_intersects(*nodemask, node_states[N_MEMORY]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | } |
| 169 | |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 170 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
| 171 | { |
Bob Liu | 6d55629 | 2010-05-24 14:31:59 -0700 | [diff] [blame] | 172 | return pol->flags & MPOL_MODE_FLAGS; |
David Rientjes | 4c50bc0 | 2008-04-28 02:12:30 -0700 | [diff] [blame] | 173 | } |
| 174 | |
| 175 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, |
| 176 | const nodemask_t *rel) |
| 177 | { |
| 178 | nodemask_t tmp; |
| 179 | nodes_fold(tmp, *orig, nodes_weight(*rel)); |
| 180 | nodes_onto(*ret, tmp, *rel); |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 181 | } |
| 182 | |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 183 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
| 184 | { |
| 185 | if (nodes_empty(*nodes)) |
| 186 | return -EINVAL; |
| 187 | pol->v.nodes = *nodes; |
| 188 | return 0; |
| 189 | } |
| 190 | |
| 191 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) |
| 192 | { |
| 193 | if (!nodes) |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 194 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 195 | else if (nodes_empty(*nodes)) |
| 196 | return -EINVAL; /* no allowed nodes */ |
| 197 | else |
| 198 | pol->v.preferred_node = first_node(*nodes); |
| 199 | return 0; |
| 200 | } |
| 201 | |
| 202 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) |
| 203 | { |
| 204 | if (!is_valid_nodemask(nodes)) |
| 205 | return -EINVAL; |
| 206 | pol->v.nodes = *nodes; |
| 207 | return 0; |
| 208 | } |
| 209 | |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 210 | /* |
| 211 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if |
| 212 | * any, for the new policy. mpol_new() has already validated the nodes |
| 213 | * parameter with respect to the policy mode and flags. But, we need to |
| 214 | * handle an empty nodemask with MPOL_PREFERRED here. |
| 215 | * |
| 216 | * Must be called holding task's alloc_lock to protect task's mems_allowed |
| 217 | * and mempolicy. May also be called holding the mmap_semaphore for write. |
| 218 | */ |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 219 | static int mpol_set_nodemask(struct mempolicy *pol, |
| 220 | const nodemask_t *nodes, struct nodemask_scratch *nsc) |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 221 | { |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 222 | int ret; |
| 223 | |
| 224 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ |
| 225 | if (pol == NULL) |
| 226 | return 0; |
Lai Jiangshan | 01f13bd | 2012-12-12 13:51:33 -0800 | [diff] [blame] | 227 | /* Check N_MEMORY */ |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 228 | nodes_and(nsc->mask1, |
Lai Jiangshan | 01f13bd | 2012-12-12 13:51:33 -0800 | [diff] [blame] | 229 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 230 | |
| 231 | VM_BUG_ON(!nodes); |
| 232 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) |
| 233 | nodes = NULL; /* explicit local allocation */ |
| 234 | else { |
| 235 | if (pol->flags & MPOL_F_RELATIVE_NODES) |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 236 | mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 237 | else |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 238 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
| 239 | |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 240 | if (mpol_store_user_nodemask(pol)) |
| 241 | pol->w.user_nodemask = *nodes; |
| 242 | else |
| 243 | pol->w.cpuset_mems_allowed = |
| 244 | cpuset_current_mems_allowed; |
| 245 | } |
| 246 | |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 247 | if (nodes) |
| 248 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); |
| 249 | else |
| 250 | ret = mpol_ops[pol->mode].create(pol, NULL); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 251 | return ret; |
| 252 | } |
| 253 | |
| 254 | /* |
| 255 | * This function just creates a new policy, does some check and simple |
| 256 | * initialization. You must invoke mpol_set_nodemask() to set nodes. |
| 257 | */ |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 258 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
| 259 | nodemask_t *nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | { |
| 261 | struct mempolicy *policy; |
| 262 | |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 263 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
David Rientjes | 00ef2d2 | 2013-02-22 16:35:36 -0800 | [diff] [blame] | 264 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 265 | |
David Rientjes | 3e1f064 | 2008-04-28 02:12:34 -0700 | [diff] [blame] | 266 | if (mode == MPOL_DEFAULT) { |
| 267 | if (nodes && !nodes_empty(*nodes)) |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 268 | return ERR_PTR(-EINVAL); |
Lee Schermerhorn | d3a7103 | 2012-10-25 14:16:29 +0200 | [diff] [blame] | 269 | return NULL; |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 270 | } |
David Rientjes | 3e1f064 | 2008-04-28 02:12:34 -0700 | [diff] [blame] | 271 | VM_BUG_ON(!nodes); |
| 272 | |
| 273 | /* |
| 274 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or |
| 275 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). |
| 276 | * All other modes require a valid pointer to a non-empty nodemask. |
| 277 | */ |
| 278 | if (mode == MPOL_PREFERRED) { |
| 279 | if (nodes_empty(*nodes)) { |
| 280 | if (((flags & MPOL_F_STATIC_NODES) || |
| 281 | (flags & MPOL_F_RELATIVE_NODES))) |
| 282 | return ERR_PTR(-EINVAL); |
David Rientjes | 3e1f064 | 2008-04-28 02:12:34 -0700 | [diff] [blame] | 283 | } |
Peter Zijlstra | 479e280 | 2012-10-25 14:16:28 +0200 | [diff] [blame] | 284 | } else if (mode == MPOL_LOCAL) { |
| 285 | if (!nodes_empty(*nodes)) |
| 286 | return ERR_PTR(-EINVAL); |
| 287 | mode = MPOL_PREFERRED; |
David Rientjes | 3e1f064 | 2008-04-28 02:12:34 -0700 | [diff] [blame] | 288 | } else if (nodes_empty(*nodes)) |
| 289 | return ERR_PTR(-EINVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
| 291 | if (!policy) |
| 292 | return ERR_PTR(-ENOMEM); |
| 293 | atomic_set(&policy->refcnt, 1); |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 294 | policy->mode = mode; |
David Rientjes | 3e1f064 | 2008-04-28 02:12:34 -0700 | [diff] [blame] | 295 | policy->flags = flags; |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 296 | |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 297 | return policy; |
| 298 | } |
| 299 | |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 300 | /* Slow path of a mpol destructor. */ |
| 301 | void __mpol_put(struct mempolicy *p) |
| 302 | { |
| 303 | if (!atomic_dec_and_test(&p->refcnt)) |
| 304 | return; |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 305 | kmem_cache_free(policy_cache, p); |
| 306 | } |
| 307 | |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 308 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes, |
| 309 | enum mpol_rebind_step step) |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 310 | { |
| 311 | } |
| 312 | |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 313 | /* |
| 314 | * step: |
| 315 | * MPOL_REBIND_ONCE - do rebind work at once |
| 316 | * MPOL_REBIND_STEP1 - set all the newly nodes |
| 317 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes |
| 318 | */ |
| 319 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes, |
| 320 | enum mpol_rebind_step step) |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 321 | { |
| 322 | nodemask_t tmp; |
| 323 | |
| 324 | if (pol->flags & MPOL_F_STATIC_NODES) |
| 325 | nodes_and(tmp, pol->w.user_nodemask, *nodes); |
| 326 | else if (pol->flags & MPOL_F_RELATIVE_NODES) |
| 327 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); |
| 328 | else { |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 329 | /* |
| 330 | * if step == 1, we use ->w.cpuset_mems_allowed to cache the |
| 331 | * result |
| 332 | */ |
| 333 | if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) { |
| 334 | nodes_remap(tmp, pol->v.nodes, |
| 335 | pol->w.cpuset_mems_allowed, *nodes); |
| 336 | pol->w.cpuset_mems_allowed = step ? tmp : *nodes; |
| 337 | } else if (step == MPOL_REBIND_STEP2) { |
| 338 | tmp = pol->w.cpuset_mems_allowed; |
| 339 | pol->w.cpuset_mems_allowed = *nodes; |
| 340 | } else |
| 341 | BUG(); |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 342 | } |
| 343 | |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 344 | if (nodes_empty(tmp)) |
| 345 | tmp = *nodes; |
| 346 | |
| 347 | if (step == MPOL_REBIND_STEP1) |
| 348 | nodes_or(pol->v.nodes, pol->v.nodes, tmp); |
| 349 | else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2) |
| 350 | pol->v.nodes = tmp; |
| 351 | else |
| 352 | BUG(); |
| 353 | |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 354 | if (!node_isset(current->il_next, tmp)) { |
| 355 | current->il_next = next_node(current->il_next, tmp); |
| 356 | if (current->il_next >= MAX_NUMNODES) |
| 357 | current->il_next = first_node(tmp); |
| 358 | if (current->il_next >= MAX_NUMNODES) |
| 359 | current->il_next = numa_node_id(); |
| 360 | } |
| 361 | } |
| 362 | |
| 363 | static void mpol_rebind_preferred(struct mempolicy *pol, |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 364 | const nodemask_t *nodes, |
| 365 | enum mpol_rebind_step step) |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 366 | { |
| 367 | nodemask_t tmp; |
| 368 | |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 369 | if (pol->flags & MPOL_F_STATIC_NODES) { |
| 370 | int node = first_node(pol->w.user_nodemask); |
| 371 | |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 372 | if (node_isset(node, *nodes)) { |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 373 | pol->v.preferred_node = node; |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 374 | pol->flags &= ~MPOL_F_LOCAL; |
| 375 | } else |
| 376 | pol->flags |= MPOL_F_LOCAL; |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 377 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
| 378 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); |
| 379 | pol->v.preferred_node = first_node(tmp); |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 380 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 381 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
| 382 | pol->w.cpuset_mems_allowed, |
| 383 | *nodes); |
| 384 | pol->w.cpuset_mems_allowed = *nodes; |
| 385 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | } |
| 387 | |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 388 | /* |
| 389 | * mpol_rebind_policy - Migrate a policy to a different set of nodes |
| 390 | * |
| 391 | * If read-side task has no lock to protect task->mempolicy, write-side |
| 392 | * task will rebind the task->mempolicy by two step. The first step is |
| 393 | * setting all the newly nodes, and the second step is cleaning all the |
| 394 | * disallowed nodes. In this way, we can avoid finding no node to alloc |
| 395 | * page. |
| 396 | * If we have a lock to protect task->mempolicy in read-side, we do |
| 397 | * rebind directly. |
| 398 | * |
| 399 | * step: |
| 400 | * MPOL_REBIND_ONCE - do rebind work at once |
| 401 | * MPOL_REBIND_STEP1 - set all the newly nodes |
| 402 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes |
| 403 | */ |
| 404 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask, |
| 405 | enum mpol_rebind_step step) |
David Rientjes | 1d0d268 | 2008-04-28 02:12:32 -0700 | [diff] [blame] | 406 | { |
David Rientjes | 1d0d268 | 2008-04-28 02:12:32 -0700 | [diff] [blame] | 407 | if (!pol) |
| 408 | return; |
Wang Sheng-Hui | 89c522c | 2012-05-29 15:06:16 -0700 | [diff] [blame] | 409 | if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE && |
David Rientjes | 1d0d268 | 2008-04-28 02:12:32 -0700 | [diff] [blame] | 410 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
| 411 | return; |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 412 | |
| 413 | if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING)) |
| 414 | return; |
| 415 | |
| 416 | if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING)) |
| 417 | BUG(); |
| 418 | |
| 419 | if (step == MPOL_REBIND_STEP1) |
| 420 | pol->flags |= MPOL_F_REBINDING; |
| 421 | else if (step == MPOL_REBIND_STEP2) |
| 422 | pol->flags &= ~MPOL_F_REBINDING; |
| 423 | else if (step >= MPOL_REBIND_NSTEP) |
| 424 | BUG(); |
| 425 | |
| 426 | mpol_ops[pol->mode].rebind(pol, newmask, step); |
David Rientjes | 1d0d268 | 2008-04-28 02:12:32 -0700 | [diff] [blame] | 427 | } |
| 428 | |
| 429 | /* |
| 430 | * Wrapper for mpol_rebind_policy() that just requires task |
| 431 | * pointer, and updates task mempolicy. |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 432 | * |
| 433 | * Called with task's alloc_lock held. |
David Rientjes | 1d0d268 | 2008-04-28 02:12:32 -0700 | [diff] [blame] | 434 | */ |
| 435 | |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 436 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new, |
| 437 | enum mpol_rebind_step step) |
David Rientjes | 1d0d268 | 2008-04-28 02:12:32 -0700 | [diff] [blame] | 438 | { |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 439 | mpol_rebind_policy(tsk->mempolicy, new, step); |
David Rientjes | 1d0d268 | 2008-04-28 02:12:32 -0700 | [diff] [blame] | 440 | } |
| 441 | |
| 442 | /* |
| 443 | * Rebind each vma in mm to new nodemask. |
| 444 | * |
| 445 | * Call holding a reference to mm. Takes mm->mmap_sem during call. |
| 446 | */ |
| 447 | |
| 448 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) |
| 449 | { |
| 450 | struct vm_area_struct *vma; |
| 451 | |
| 452 | down_write(&mm->mmap_sem); |
| 453 | for (vma = mm->mmap; vma; vma = vma->vm_next) |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 454 | mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE); |
David Rientjes | 1d0d268 | 2008-04-28 02:12:32 -0700 | [diff] [blame] | 455 | up_write(&mm->mmap_sem); |
| 456 | } |
| 457 | |
David Rientjes | 3701294 | 2008-04-28 02:12:33 -0700 | [diff] [blame] | 458 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
| 459 | [MPOL_DEFAULT] = { |
| 460 | .rebind = mpol_rebind_default, |
| 461 | }, |
| 462 | [MPOL_INTERLEAVE] = { |
| 463 | .create = mpol_new_interleave, |
| 464 | .rebind = mpol_rebind_nodemask, |
| 465 | }, |
| 466 | [MPOL_PREFERRED] = { |
| 467 | .create = mpol_new_preferred, |
| 468 | .rebind = mpol_rebind_preferred, |
| 469 | }, |
| 470 | [MPOL_BIND] = { |
| 471 | .create = mpol_new_bind, |
| 472 | .rebind = mpol_rebind_nodemask, |
| 473 | }, |
| 474 | }; |
| 475 | |
Christoph Lameter | fc30128 | 2006-01-18 17:42:29 -0800 | [diff] [blame] | 476 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
| 477 | unsigned long flags); |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 478 | |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 479 | /* |
| 480 | * Scan through pages checking if pages follow certain conditions, |
| 481 | * and move them to the pagelist if they do. |
| 482 | */ |
| 483 | static int queue_pages_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 484 | unsigned long addr, unsigned long end, |
| 485 | const nodemask_t *nodes, unsigned long flags, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 486 | void *private) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 487 | { |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 488 | pte_t *orig_pte; |
| 489 | pte_t *pte; |
Hugh Dickins | 705e87c | 2005-10-29 18:16:27 -0700 | [diff] [blame] | 490 | spinlock_t *ptl; |
Hugh Dickins | 941150a | 2005-06-21 17:15:06 -0700 | [diff] [blame] | 491 | |
Hugh Dickins | 705e87c | 2005-10-29 18:16:27 -0700 | [diff] [blame] | 492 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 493 | do { |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 494 | struct page *page; |
Andy Whitcroft | 25ba77c | 2006-12-06 20:33:03 -0800 | [diff] [blame] | 495 | int nid; |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 496 | |
| 497 | if (!pte_present(*pte)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | continue; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 499 | page = vm_normal_page(vma, addr, *pte); |
| 500 | if (!page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | continue; |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 502 | /* |
Hugh Dickins | 62b61f6 | 2009-12-14 17:59:33 -0800 | [diff] [blame] | 503 | * vm_normal_page() filters out zero pages, but there might |
| 504 | * still be PageReserved pages to skip, perhaps in a VDSO. |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 505 | */ |
Hugh Dickins | b79bc0a | 2013-02-22 16:35:13 -0800 | [diff] [blame] | 506 | if (PageReserved(page)) |
Christoph Lameter | f4598c8 | 2006-01-12 01:05:20 -0800 | [diff] [blame] | 507 | continue; |
Linus Torvalds | 6aab341 | 2005-11-28 14:34:23 -0800 | [diff] [blame] | 508 | nid = page_to_nid(page); |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 509 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
| 510 | continue; |
| 511 | |
Stephen Wilson | b1f72d1 | 2011-05-24 17:12:43 -0700 | [diff] [blame] | 512 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
Christoph Lameter | fc30128 | 2006-01-18 17:42:29 -0800 | [diff] [blame] | 513 | migrate_page_add(page, private, flags); |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 514 | else |
| 515 | break; |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 516 | } while (pte++, addr += PAGE_SIZE, addr != end); |
Hugh Dickins | 705e87c | 2005-10-29 18:16:27 -0700 | [diff] [blame] | 517 | pte_unmap_unlock(orig_pte, ptl); |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 518 | return addr != end; |
| 519 | } |
| 520 | |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 521 | static void queue_pages_hugetlb_pmd_range(struct vm_area_struct *vma, |
| 522 | pmd_t *pmd, const nodemask_t *nodes, unsigned long flags, |
Naoya Horiguchi | e2d8cf4 | 2013-09-11 14:22:03 -0700 | [diff] [blame] | 523 | void *private) |
| 524 | { |
| 525 | #ifdef CONFIG_HUGETLB_PAGE |
| 526 | int nid; |
| 527 | struct page *page; |
| 528 | |
| 529 | spin_lock(&vma->vm_mm->page_table_lock); |
| 530 | page = pte_page(huge_ptep_get((pte_t *)pmd)); |
| 531 | nid = page_to_nid(page); |
| 532 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
| 533 | goto unlock; |
| 534 | /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */ |
| 535 | if (flags & (MPOL_MF_MOVE_ALL) || |
| 536 | (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) |
| 537 | isolate_huge_page(page, private); |
| 538 | unlock: |
| 539 | spin_unlock(&vma->vm_mm->page_table_lock); |
| 540 | #else |
| 541 | BUG(); |
| 542 | #endif |
| 543 | } |
| 544 | |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 545 | static inline int queue_pages_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 546 | unsigned long addr, unsigned long end, |
| 547 | const nodemask_t *nodes, unsigned long flags, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 548 | void *private) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 549 | { |
| 550 | pmd_t *pmd; |
| 551 | unsigned long next; |
| 552 | |
| 553 | pmd = pmd_offset(pud, addr); |
| 554 | do { |
| 555 | next = pmd_addr_end(addr, end); |
Naoya Horiguchi | e2d8cf4 | 2013-09-11 14:22:03 -0700 | [diff] [blame] | 556 | if (!pmd_present(*pmd)) |
| 557 | continue; |
| 558 | if (pmd_huge(*pmd) && is_vm_hugetlb_page(vma)) { |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 559 | queue_pages_hugetlb_pmd_range(vma, pmd, nodes, |
Naoya Horiguchi | e2d8cf4 | 2013-09-11 14:22:03 -0700 | [diff] [blame] | 560 | flags, private); |
| 561 | continue; |
| 562 | } |
Kirill A. Shutemov | e180377 | 2012-12-12 13:50:59 -0800 | [diff] [blame] | 563 | split_huge_page_pmd(vma, addr, pmd); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 564 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 565 | continue; |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 566 | if (queue_pages_pte_range(vma, pmd, addr, next, nodes, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 567 | flags, private)) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 568 | return -EIO; |
| 569 | } while (pmd++, addr = next, addr != end); |
| 570 | return 0; |
| 571 | } |
| 572 | |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 573 | static inline int queue_pages_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 574 | unsigned long addr, unsigned long end, |
| 575 | const nodemask_t *nodes, unsigned long flags, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 576 | void *private) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 577 | { |
| 578 | pud_t *pud; |
| 579 | unsigned long next; |
| 580 | |
| 581 | pud = pud_offset(pgd, addr); |
| 582 | do { |
| 583 | next = pud_addr_end(addr, end); |
Naoya Horiguchi | e2d8cf4 | 2013-09-11 14:22:03 -0700 | [diff] [blame] | 584 | if (pud_huge(*pud) && is_vm_hugetlb_page(vma)) |
| 585 | continue; |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 586 | if (pud_none_or_clear_bad(pud)) |
| 587 | continue; |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 588 | if (queue_pages_pmd_range(vma, pud, addr, next, nodes, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 589 | flags, private)) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 590 | return -EIO; |
| 591 | } while (pud++, addr = next, addr != end); |
| 592 | return 0; |
| 593 | } |
| 594 | |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 595 | static inline int queue_pages_pgd_range(struct vm_area_struct *vma, |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 596 | unsigned long addr, unsigned long end, |
| 597 | const nodemask_t *nodes, unsigned long flags, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 598 | void *private) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 599 | { |
| 600 | pgd_t *pgd; |
| 601 | unsigned long next; |
| 602 | |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 603 | pgd = pgd_offset(vma->vm_mm, addr); |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 604 | do { |
| 605 | next = pgd_addr_end(addr, end); |
| 606 | if (pgd_none_or_clear_bad(pgd)) |
| 607 | continue; |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 608 | if (queue_pages_pud_range(vma, pgd, addr, next, nodes, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 609 | flags, private)) |
Hugh Dickins | 91612e0 | 2005-06-21 17:15:07 -0700 | [diff] [blame] | 610 | return -EIO; |
| 611 | } while (pgd++, addr = next, addr != end); |
| 612 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | } |
| 614 | |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 615 | #ifdef CONFIG_ARCH_USES_NUMA_PROT_NONE |
| 616 | /* |
Mel Gorman | 4b10e7d | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 617 | * This is used to mark a range of virtual addresses to be inaccessible. |
| 618 | * These are later cleared by a NUMA hinting fault. Depending on these |
| 619 | * faults, pages may be migrated for better NUMA placement. |
| 620 | * |
| 621 | * This is assuming that NUMA faults are handled using PROT_NONE. If |
| 622 | * an architecture makes a different choice, it will need further |
| 623 | * changes to the core. |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 624 | */ |
Mel Gorman | 4b10e7d | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 625 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
| 626 | unsigned long addr, unsigned long end) |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 627 | { |
Mel Gorman | 4b10e7d | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 628 | int nr_updated; |
| 629 | BUILD_BUG_ON(_PAGE_NUMA != _PAGE_PROTNONE); |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 630 | |
Mel Gorman | 4b10e7d | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 631 | nr_updated = change_protection(vma, addr, end, vma->vm_page_prot, 0, 1); |
Mel Gorman | 03c5a6e | 2012-11-02 14:52:48 +0000 | [diff] [blame] | 632 | if (nr_updated) |
| 633 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 634 | |
Mel Gorman | 4b10e7d | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 635 | return nr_updated; |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 636 | } |
| 637 | #else |
| 638 | static unsigned long change_prot_numa(struct vm_area_struct *vma, |
| 639 | unsigned long addr, unsigned long end) |
| 640 | { |
| 641 | return 0; |
| 642 | } |
| 643 | #endif /* CONFIG_ARCH_USES_NUMA_PROT_NONE */ |
| 644 | |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 645 | /* |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 646 | * Walk through page tables and collect pages to be migrated. |
| 647 | * |
| 648 | * If pages found in a given range are on a set of nodes (determined by |
| 649 | * @nodes and @flags,) it's isolated and queued to the pagelist which is |
| 650 | * passed via @private.) |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 651 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | static struct vm_area_struct * |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 653 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 654 | const nodemask_t *nodes, unsigned long flags, void *private) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | { |
| 656 | int err; |
| 657 | struct vm_area_struct *first, *vma, *prev; |
| 658 | |
Nick Piggin | 053837f | 2006-01-18 17:42:27 -0800 | [diff] [blame] | 659 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 660 | first = find_vma(mm, start); |
| 661 | if (!first) |
| 662 | return ERR_PTR(-EFAULT); |
| 663 | prev = NULL; |
| 664 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 665 | unsigned long endvma = vma->vm_end; |
| 666 | |
| 667 | if (endvma > end) |
| 668 | endvma = end; |
| 669 | if (vma->vm_start > start) |
| 670 | start = vma->vm_start; |
| 671 | |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 672 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
| 673 | if (!vma->vm_next && vma->vm_end < end) |
| 674 | return ERR_PTR(-EFAULT); |
| 675 | if (prev && prev->vm_end < vma->vm_start) |
| 676 | return ERR_PTR(-EFAULT); |
| 677 | } |
Christoph Lameter | dc9aa5b | 2006-01-08 01:00:50 -0800 | [diff] [blame] | 678 | |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 679 | if (flags & MPOL_MF_LAZY) { |
| 680 | change_prot_numa(vma, start, endvma); |
| 681 | goto next; |
| 682 | } |
| 683 | |
| 684 | if ((flags & MPOL_MF_STRICT) || |
| 685 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && |
| 686 | vma_migratable(vma))) { |
| 687 | |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 688 | err = queue_pages_pgd_range(vma, start, endvma, nodes, |
Christoph Lameter | 38e3586 | 2006-01-08 01:01:01 -0800 | [diff] [blame] | 689 | flags, private); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | if (err) { |
| 691 | first = ERR_PTR(err); |
| 692 | break; |
| 693 | } |
| 694 | } |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 695 | next: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | prev = vma; |
| 697 | } |
| 698 | return first; |
| 699 | } |
| 700 | |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 701 | /* |
| 702 | * Apply policy to a single VMA |
| 703 | * This must be called with the mmap_sem held for writing. |
| 704 | */ |
| 705 | static int vma_replace_policy(struct vm_area_struct *vma, |
| 706 | struct mempolicy *pol) |
KOSAKI Motohiro | 8d34694c1 | 2012-10-08 16:29:14 -0700 | [diff] [blame] | 707 | { |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 708 | int err; |
| 709 | struct mempolicy *old; |
| 710 | struct mempolicy *new; |
KOSAKI Motohiro | 8d34694c1 | 2012-10-08 16:29:14 -0700 | [diff] [blame] | 711 | |
| 712 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
| 713 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
| 714 | vma->vm_ops, vma->vm_file, |
| 715 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); |
| 716 | |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 717 | new = mpol_dup(pol); |
| 718 | if (IS_ERR(new)) |
| 719 | return PTR_ERR(new); |
| 720 | |
| 721 | if (vma->vm_ops && vma->vm_ops->set_policy) { |
KOSAKI Motohiro | 8d34694c1 | 2012-10-08 16:29:14 -0700 | [diff] [blame] | 722 | err = vma->vm_ops->set_policy(vma, new); |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 723 | if (err) |
| 724 | goto err_out; |
KOSAKI Motohiro | 8d34694c1 | 2012-10-08 16:29:14 -0700 | [diff] [blame] | 725 | } |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 726 | |
| 727 | old = vma->vm_policy; |
| 728 | vma->vm_policy = new; /* protected by mmap_sem */ |
| 729 | mpol_put(old); |
| 730 | |
| 731 | return 0; |
| 732 | err_out: |
| 733 | mpol_put(new); |
KOSAKI Motohiro | 8d34694c1 | 2012-10-08 16:29:14 -0700 | [diff] [blame] | 734 | return err; |
| 735 | } |
| 736 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 737 | /* Step 2: apply policy to a range and do splits. */ |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 738 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
| 739 | unsigned long end, struct mempolicy *new_pol) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | { |
| 741 | struct vm_area_struct *next; |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 742 | struct vm_area_struct *prev; |
| 743 | struct vm_area_struct *vma; |
| 744 | int err = 0; |
KOSAKI Motohiro | e26a511 | 2011-12-28 15:57:11 -0800 | [diff] [blame] | 745 | pgoff_t pgoff; |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 746 | unsigned long vmstart; |
| 747 | unsigned long vmend; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 748 | |
Linus Torvalds | 097d591 | 2012-03-06 18:23:36 -0800 | [diff] [blame] | 749 | vma = find_vma(mm, start); |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 750 | if (!vma || vma->vm_start > start) |
| 751 | return -EFAULT; |
| 752 | |
Linus Torvalds | 097d591 | 2012-03-06 18:23:36 -0800 | [diff] [blame] | 753 | prev = vma->vm_prev; |
KOSAKI Motohiro | e26a511 | 2011-12-28 15:57:11 -0800 | [diff] [blame] | 754 | if (start > vma->vm_start) |
| 755 | prev = vma; |
| 756 | |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 757 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | next = vma->vm_next; |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 759 | vmstart = max(start, vma->vm_start); |
| 760 | vmend = min(end, vma->vm_end); |
| 761 | |
KOSAKI Motohiro | e26a511 | 2011-12-28 15:57:11 -0800 | [diff] [blame] | 762 | if (mpol_equal(vma_policy(vma), new_pol)) |
| 763 | continue; |
| 764 | |
| 765 | pgoff = vma->vm_pgoff + |
| 766 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 767 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
KOSAKI Motohiro | e26a511 | 2011-12-28 15:57:11 -0800 | [diff] [blame] | 768 | vma->anon_vma, vma->vm_file, pgoff, |
Caspar Zhang | 8aacc9f | 2011-09-14 16:20:58 -0700 | [diff] [blame] | 769 | new_pol); |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 770 | if (prev) { |
| 771 | vma = prev; |
| 772 | next = vma->vm_next; |
Oleg Nesterov | 3964acd | 2013-07-31 13:53:28 -0700 | [diff] [blame] | 773 | if (mpol_equal(vma_policy(vma), new_pol)) |
| 774 | continue; |
| 775 | /* vma_merge() joined vma && vma->next, case 8 */ |
| 776 | goto replace; |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 777 | } |
| 778 | if (vma->vm_start != vmstart) { |
| 779 | err = split_vma(vma->vm_mm, vma, vmstart, 1); |
| 780 | if (err) |
| 781 | goto out; |
| 782 | } |
| 783 | if (vma->vm_end != vmend) { |
| 784 | err = split_vma(vma->vm_mm, vma, vmend, 0); |
| 785 | if (err) |
| 786 | goto out; |
| 787 | } |
Oleg Nesterov | 3964acd | 2013-07-31 13:53:28 -0700 | [diff] [blame] | 788 | replace: |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 789 | err = vma_replace_policy(vma, new_pol); |
KOSAKI Motohiro | 8d34694c1 | 2012-10-08 16:29:14 -0700 | [diff] [blame] | 790 | if (err) |
| 791 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | } |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 793 | |
| 794 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | return err; |
| 796 | } |
| 797 | |
Paul Jackson | c61afb1 | 2006-03-24 03:16:08 -0800 | [diff] [blame] | 798 | /* |
| 799 | * Update task->flags PF_MEMPOLICY bit: set iff non-default |
| 800 | * mempolicy. Allows more rapid checking of this (combined perhaps |
| 801 | * with other PF_* flag bits) on memory allocation hot code paths. |
| 802 | * |
| 803 | * If called from outside this file, the task 'p' should -only- be |
| 804 | * a newly forked child not yet visible on the task list, because |
| 805 | * manipulating the task flags of a visible task is not safe. |
| 806 | * |
| 807 | * The above limitation is why this routine has the funny name |
| 808 | * mpol_fix_fork_child_flag(). |
| 809 | * |
| 810 | * It is also safe to call this with a task pointer of current, |
| 811 | * which the static wrapper mpol_set_task_struct_flag() does, |
| 812 | * for use within this file. |
| 813 | */ |
| 814 | |
| 815 | void mpol_fix_fork_child_flag(struct task_struct *p) |
| 816 | { |
| 817 | if (p->mempolicy) |
| 818 | p->flags |= PF_MEMPOLICY; |
| 819 | else |
| 820 | p->flags &= ~PF_MEMPOLICY; |
| 821 | } |
| 822 | |
| 823 | static void mpol_set_task_struct_flag(void) |
| 824 | { |
| 825 | mpol_fix_fork_child_flag(current); |
| 826 | } |
| 827 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | /* Set the process memory policy */ |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 829 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
| 830 | nodemask_t *nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 831 | { |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 832 | struct mempolicy *new, *old; |
Lee Schermerhorn | f4e53d9 | 2008-04-28 02:13:10 -0700 | [diff] [blame] | 833 | struct mm_struct *mm = current->mm; |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 834 | NODEMASK_SCRATCH(scratch); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 835 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 836 | |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 837 | if (!scratch) |
| 838 | return -ENOMEM; |
Lee Schermerhorn | f4e53d9 | 2008-04-28 02:13:10 -0700 | [diff] [blame] | 839 | |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 840 | new = mpol_new(mode, flags, nodes); |
| 841 | if (IS_ERR(new)) { |
| 842 | ret = PTR_ERR(new); |
| 843 | goto out; |
| 844 | } |
Lee Schermerhorn | f4e53d9 | 2008-04-28 02:13:10 -0700 | [diff] [blame] | 845 | /* |
| 846 | * prevent changing our mempolicy while show_numa_maps() |
| 847 | * is using it. |
| 848 | * Note: do_set_mempolicy() can be called at init time |
| 849 | * with no 'mm'. |
| 850 | */ |
| 851 | if (mm) |
| 852 | down_write(&mm->mmap_sem); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 853 | task_lock(current); |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 854 | ret = mpol_set_nodemask(new, nodes, scratch); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 855 | if (ret) { |
| 856 | task_unlock(current); |
| 857 | if (mm) |
| 858 | up_write(&mm->mmap_sem); |
| 859 | mpol_put(new); |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 860 | goto out; |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 861 | } |
| 862 | old = current->mempolicy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | current->mempolicy = new; |
Paul Jackson | c61afb1 | 2006-03-24 03:16:08 -0800 | [diff] [blame] | 864 | mpol_set_task_struct_flag(); |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 865 | if (new && new->mode == MPOL_INTERLEAVE && |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 866 | nodes_weight(new->v.nodes)) |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 867 | current->il_next = first_node(new->v.nodes); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 868 | task_unlock(current); |
Lee Schermerhorn | f4e53d9 | 2008-04-28 02:13:10 -0700 | [diff] [blame] | 869 | if (mm) |
| 870 | up_write(&mm->mmap_sem); |
| 871 | |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 872 | mpol_put(old); |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 873 | ret = 0; |
| 874 | out: |
| 875 | NODEMASK_SCRATCH_FREE(scratch); |
| 876 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 877 | } |
| 878 | |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 879 | /* |
| 880 | * Return nodemask for policy for get_mempolicy() query |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 881 | * |
| 882 | * Called with task's alloc_lock held |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 883 | */ |
| 884 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | { |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 886 | nodes_clear(*nodes); |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 887 | if (p == &default_policy) |
| 888 | return; |
| 889 | |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 890 | switch (p->mode) { |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 891 | case MPOL_BIND: |
| 892 | /* Fall through */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | case MPOL_INTERLEAVE: |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 894 | *nodes = p->v.nodes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 895 | break; |
| 896 | case MPOL_PREFERRED: |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 897 | if (!(p->flags & MPOL_F_LOCAL)) |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 898 | node_set(p->v.preferred_node, *nodes); |
Lee Schermerhorn | 53f2556 | 2008-04-28 02:13:20 -0700 | [diff] [blame] | 899 | /* else return empty node mask for local allocation */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 900 | break; |
| 901 | default: |
| 902 | BUG(); |
| 903 | } |
| 904 | } |
| 905 | |
| 906 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
| 907 | { |
| 908 | struct page *p; |
| 909 | int err; |
| 910 | |
| 911 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); |
| 912 | if (err >= 0) { |
| 913 | err = page_to_nid(p); |
| 914 | put_page(p); |
| 915 | } |
| 916 | return err; |
| 917 | } |
| 918 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | /* Retrieve NUMA policy */ |
Adrian Bunk | dbcb0f1 | 2007-10-16 01:26:26 -0700 | [diff] [blame] | 920 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
| 921 | unsigned long addr, unsigned long flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 922 | { |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 923 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | struct mm_struct *mm = current->mm; |
| 925 | struct vm_area_struct *vma = NULL; |
| 926 | struct mempolicy *pol = current->mempolicy; |
| 927 | |
Lee Schermerhorn | 754af6f | 2007-10-16 01:24:51 -0700 | [diff] [blame] | 928 | if (flags & |
| 929 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 930 | return -EINVAL; |
Lee Schermerhorn | 754af6f | 2007-10-16 01:24:51 -0700 | [diff] [blame] | 931 | |
| 932 | if (flags & MPOL_F_MEMS_ALLOWED) { |
| 933 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) |
| 934 | return -EINVAL; |
| 935 | *policy = 0; /* just so it's initialized */ |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 936 | task_lock(current); |
Lee Schermerhorn | 754af6f | 2007-10-16 01:24:51 -0700 | [diff] [blame] | 937 | *nmask = cpuset_current_mems_allowed; |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 938 | task_unlock(current); |
Lee Schermerhorn | 754af6f | 2007-10-16 01:24:51 -0700 | [diff] [blame] | 939 | return 0; |
| 940 | } |
| 941 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 942 | if (flags & MPOL_F_ADDR) { |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 943 | /* |
| 944 | * Do NOT fall back to task policy if the |
| 945 | * vma/shared policy at addr is NULL. We |
| 946 | * want to return MPOL_DEFAULT in this case. |
| 947 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 948 | down_read(&mm->mmap_sem); |
| 949 | vma = find_vma_intersection(mm, addr, addr+1); |
| 950 | if (!vma) { |
| 951 | up_read(&mm->mmap_sem); |
| 952 | return -EFAULT; |
| 953 | } |
| 954 | if (vma->vm_ops && vma->vm_ops->get_policy) |
| 955 | pol = vma->vm_ops->get_policy(vma, addr); |
| 956 | else |
| 957 | pol = vma->vm_policy; |
| 958 | } else if (addr) |
| 959 | return -EINVAL; |
| 960 | |
| 961 | if (!pol) |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 962 | pol = &default_policy; /* indicates default behavior */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 963 | |
| 964 | if (flags & MPOL_F_NODE) { |
| 965 | if (flags & MPOL_F_ADDR) { |
| 966 | err = lookup_node(mm, addr); |
| 967 | if (err < 0) |
| 968 | goto out; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 969 | *policy = err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 970 | } else if (pol == current->mempolicy && |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 971 | pol->mode == MPOL_INTERLEAVE) { |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 972 | *policy = current->il_next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | } else { |
| 974 | err = -EINVAL; |
| 975 | goto out; |
| 976 | } |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 977 | } else { |
| 978 | *policy = pol == &default_policy ? MPOL_DEFAULT : |
| 979 | pol->mode; |
David Rientjes | d79df63 | 2008-07-04 12:24:13 -0700 | [diff] [blame] | 980 | /* |
| 981 | * Internal mempolicy flags must be masked off before exposing |
| 982 | * the policy to userspace. |
| 983 | */ |
| 984 | *policy |= (pol->flags & MPOL_MODE_FLAGS); |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 985 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 986 | |
| 987 | if (vma) { |
| 988 | up_read(¤t->mm->mmap_sem); |
| 989 | vma = NULL; |
| 990 | } |
| 991 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 992 | err = 0; |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 993 | if (nmask) { |
Lee Schermerhorn | c6b6ef8 | 2010-03-23 13:35:41 -0700 | [diff] [blame] | 994 | if (mpol_store_user_nodemask(pol)) { |
| 995 | *nmask = pol->w.user_nodemask; |
| 996 | } else { |
| 997 | task_lock(current); |
| 998 | get_policy_nodemask(pol, nmask); |
| 999 | task_unlock(current); |
| 1000 | } |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 1001 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | |
| 1003 | out: |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1004 | mpol_cond_put(pol); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | if (vma) |
| 1006 | up_read(¤t->mm->mmap_sem); |
| 1007 | return err; |
| 1008 | } |
| 1009 | |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1010 | #ifdef CONFIG_MIGRATION |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1011 | /* |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1012 | * page migration |
| 1013 | */ |
Christoph Lameter | fc30128 | 2006-01-18 17:42:29 -0800 | [diff] [blame] | 1014 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
| 1015 | unsigned long flags) |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1016 | { |
| 1017 | /* |
Christoph Lameter | fc30128 | 2006-01-18 17:42:29 -0800 | [diff] [blame] | 1018 | * Avoid migrating a page that is shared with others. |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1019 | */ |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1020 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
| 1021 | if (!isolate_lru_page(page)) { |
| 1022 | list_add_tail(&page->lru, pagelist); |
KOSAKI Motohiro | 6d9c285 | 2009-12-14 17:58:11 -0800 | [diff] [blame] | 1023 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
| 1024 | page_is_file_cache(page)); |
Nick Piggin | 62695a8 | 2008-10-18 20:26:09 -0700 | [diff] [blame] | 1025 | } |
| 1026 | } |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1027 | } |
| 1028 | |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1029 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1030 | { |
Naoya Horiguchi | e2d8cf4 | 2013-09-11 14:22:03 -0700 | [diff] [blame] | 1031 | if (PageHuge(page)) |
| 1032 | return alloc_huge_page_node(page_hstate(compound_head(page)), |
| 1033 | node); |
| 1034 | else |
| 1035 | return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1036 | } |
| 1037 | |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1038 | /* |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1039 | * Migrate pages from one node to a target node. |
| 1040 | * Returns error or the number of pages not migrated. |
| 1041 | */ |
Adrian Bunk | dbcb0f1 | 2007-10-16 01:26:26 -0700 | [diff] [blame] | 1042 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
| 1043 | int flags) |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1044 | { |
| 1045 | nodemask_t nmask; |
| 1046 | LIST_HEAD(pagelist); |
| 1047 | int err = 0; |
| 1048 | |
| 1049 | nodes_clear(nmask); |
| 1050 | node_set(source, nmask); |
| 1051 | |
Minchan Kim | 0827080 | 2012-10-08 16:33:38 -0700 | [diff] [blame] | 1052 | /* |
| 1053 | * This does not "check" the range but isolates all pages that |
| 1054 | * need migration. Between passing in the full user address |
| 1055 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. |
| 1056 | */ |
| 1057 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 1058 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1059 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
| 1060 | |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1061 | if (!list_empty(&pagelist)) { |
Mel Gorman | 7f0f249 | 2011-01-13 15:45:58 -0800 | [diff] [blame] | 1062 | err = migrate_pages(&pagelist, new_node_page, dest, |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 1063 | MIGRATE_SYNC, MR_SYSCALL); |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1064 | if (err) |
Naoya Horiguchi | e2d8cf4 | 2013-09-11 14:22:03 -0700 | [diff] [blame] | 1065 | putback_movable_pages(&pagelist); |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1066 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1067 | |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1068 | return err; |
| 1069 | } |
| 1070 | |
| 1071 | /* |
| 1072 | * Move pages between the two nodesets so as to preserve the physical |
| 1073 | * layout as much as possible. |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1074 | * |
| 1075 | * Returns the number of page that could not be moved. |
| 1076 | */ |
Andrew Morton | 0ce72d4 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1077 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
| 1078 | const nodemask_t *to, int flags) |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1079 | { |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1080 | int busy = 0; |
Christoph Lameter | 0aedadf | 2008-11-06 12:53:30 -0800 | [diff] [blame] | 1081 | int err; |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1082 | nodemask_t tmp; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1083 | |
Christoph Lameter | 0aedadf | 2008-11-06 12:53:30 -0800 | [diff] [blame] | 1084 | err = migrate_prep(); |
| 1085 | if (err) |
| 1086 | return err; |
| 1087 | |
Lee Schermerhorn | 53f2556 | 2008-04-28 02:13:20 -0700 | [diff] [blame] | 1088 | down_read(&mm->mmap_sem); |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1089 | |
Andrew Morton | 0ce72d4 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1090 | err = migrate_vmas(mm, from, to, flags); |
Christoph Lameter | 7b2259b | 2006-06-25 05:46:48 -0700 | [diff] [blame] | 1091 | if (err) |
| 1092 | goto out; |
| 1093 | |
KOSAKI Motohiro | da0aa13 | 2010-03-05 13:41:59 -0800 | [diff] [blame] | 1094 | /* |
| 1095 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' |
| 1096 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' |
| 1097 | * bit in 'tmp', and return that <source, dest> pair for migration. |
| 1098 | * The pair of nodemasks 'to' and 'from' define the map. |
| 1099 | * |
| 1100 | * If no pair of bits is found that way, fallback to picking some |
| 1101 | * pair of 'source' and 'dest' bits that are not the same. If the |
| 1102 | * 'source' and 'dest' bits are the same, this represents a node |
| 1103 | * that will be migrating to itself, so no pages need move. |
| 1104 | * |
| 1105 | * If no bits are left in 'tmp', or if all remaining bits left |
| 1106 | * in 'tmp' correspond to the same bit in 'to', return false |
| 1107 | * (nothing left to migrate). |
| 1108 | * |
| 1109 | * This lets us pick a pair of nodes to migrate between, such that |
| 1110 | * if possible the dest node is not already occupied by some other |
| 1111 | * source node, minimizing the risk of overloading the memory on a |
| 1112 | * node that would happen if we migrated incoming memory to a node |
| 1113 | * before migrating outgoing memory source that same node. |
| 1114 | * |
| 1115 | * A single scan of tmp is sufficient. As we go, we remember the |
| 1116 | * most recent <s, d> pair that moved (s != d). If we find a pair |
| 1117 | * that not only moved, but what's better, moved to an empty slot |
| 1118 | * (d is not set in tmp), then we break out then, with that pair. |
Justin P. Mattock | ae0e47f | 2011-03-01 15:06:02 +0100 | [diff] [blame] | 1119 | * Otherwise when we finish scanning from_tmp, we at least have the |
KOSAKI Motohiro | da0aa13 | 2010-03-05 13:41:59 -0800 | [diff] [blame] | 1120 | * most recent <s, d> pair that moved. If we get all the way through |
| 1121 | * the scan of tmp without finding any node that moved, much less |
| 1122 | * moved to an empty node, then there is nothing left worth migrating. |
| 1123 | */ |
Christoph Lameter | d498471 | 2006-01-08 01:00:55 -0800 | [diff] [blame] | 1124 | |
Andrew Morton | 0ce72d4 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1125 | tmp = *from; |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1126 | while (!nodes_empty(tmp)) { |
| 1127 | int s,d; |
| 1128 | int source = -1; |
| 1129 | int dest = 0; |
| 1130 | |
| 1131 | for_each_node_mask(s, tmp) { |
Larry Woodman | 4a5b18c | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1132 | |
| 1133 | /* |
| 1134 | * do_migrate_pages() tries to maintain the relative |
| 1135 | * node relationship of the pages established between |
| 1136 | * threads and memory areas. |
| 1137 | * |
| 1138 | * However if the number of source nodes is not equal to |
| 1139 | * the number of destination nodes we can not preserve |
| 1140 | * this node relative relationship. In that case, skip |
| 1141 | * copying memory from a node that is in the destination |
| 1142 | * mask. |
| 1143 | * |
| 1144 | * Example: [2,3,4] -> [3,4,5] moves everything. |
| 1145 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. |
| 1146 | */ |
| 1147 | |
Andrew Morton | 0ce72d4 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1148 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
| 1149 | (node_isset(s, *to))) |
Larry Woodman | 4a5b18c | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1150 | continue; |
| 1151 | |
Andrew Morton | 0ce72d4 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1152 | d = node_remap(s, *from, *to); |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1153 | if (s == d) |
| 1154 | continue; |
| 1155 | |
| 1156 | source = s; /* Node moved. Memorize */ |
| 1157 | dest = d; |
| 1158 | |
| 1159 | /* dest not in remaining from nodes? */ |
| 1160 | if (!node_isset(dest, tmp)) |
| 1161 | break; |
| 1162 | } |
| 1163 | if (source == -1) |
| 1164 | break; |
| 1165 | |
| 1166 | node_clear(source, tmp); |
| 1167 | err = migrate_to_node(mm, source, dest, flags); |
| 1168 | if (err > 0) |
| 1169 | busy += err; |
| 1170 | if (err < 0) |
| 1171 | break; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1172 | } |
Christoph Lameter | 7b2259b | 2006-06-25 05:46:48 -0700 | [diff] [blame] | 1173 | out: |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1174 | up_read(&mm->mmap_sem); |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1175 | if (err < 0) |
| 1176 | return err; |
| 1177 | return busy; |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1178 | |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1179 | } |
| 1180 | |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 1181 | /* |
| 1182 | * Allocate a new page for page migration based on vma policy. |
| 1183 | * Start assuming that page is mapped by vma pointed to by @private. |
| 1184 | * Search forward from there, if not. N.B., this assumes that the |
| 1185 | * list of pages handed to migrate_pages()--which is how we get here-- |
| 1186 | * is in virtual address order. |
| 1187 | */ |
Christoph Lameter | 742755a | 2006-06-23 02:03:55 -0700 | [diff] [blame] | 1188 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1189 | { |
| 1190 | struct vm_area_struct *vma = (struct vm_area_struct *)private; |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 1191 | unsigned long uninitialized_var(address); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1192 | |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 1193 | while (vma) { |
| 1194 | address = page_address_in_vma(page, vma); |
| 1195 | if (address != -EFAULT) |
| 1196 | break; |
| 1197 | vma = vma->vm_next; |
| 1198 | } |
| 1199 | |
Naoya Horiguchi | 74060e4 | 2013-09-11 14:22:06 -0700 | [diff] [blame] | 1200 | if (PageHuge(page)) |
| 1201 | return alloc_huge_page_noerr(vma, address, 1); |
Lee Schermerhorn | 3ad33b2 | 2007-11-14 16:59:10 -0800 | [diff] [blame] | 1202 | /* |
| 1203 | * if !vma, alloc_page_vma() will use task or system default policy |
| 1204 | */ |
| 1205 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1206 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1207 | #else |
| 1208 | |
| 1209 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
| 1210 | unsigned long flags) |
| 1211 | { |
| 1212 | } |
| 1213 | |
Andrew Morton | 0ce72d4 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1214 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
| 1215 | const nodemask_t *to, int flags) |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1216 | { |
| 1217 | return -ENOSYS; |
| 1218 | } |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1219 | |
Keith Owens | 6993974 | 2006-10-11 01:21:28 -0700 | [diff] [blame] | 1220 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1221 | { |
| 1222 | return NULL; |
| 1223 | } |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1224 | #endif |
| 1225 | |
Adrian Bunk | dbcb0f1 | 2007-10-16 01:26:26 -0700 | [diff] [blame] | 1226 | static long do_mbind(unsigned long start, unsigned long len, |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1227 | unsigned short mode, unsigned short mode_flags, |
| 1228 | nodemask_t *nmask, unsigned long flags) |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1229 | { |
| 1230 | struct vm_area_struct *vma; |
| 1231 | struct mm_struct *mm = current->mm; |
| 1232 | struct mempolicy *new; |
| 1233 | unsigned long end; |
| 1234 | int err; |
| 1235 | LIST_HEAD(pagelist); |
| 1236 | |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 1237 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1238 | return -EINVAL; |
Christoph Lameter | 74c0024 | 2006-03-14 19:50:21 -0800 | [diff] [blame] | 1239 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1240 | return -EPERM; |
| 1241 | |
| 1242 | if (start & ~PAGE_MASK) |
| 1243 | return -EINVAL; |
| 1244 | |
| 1245 | if (mode == MPOL_DEFAULT) |
| 1246 | flags &= ~MPOL_MF_STRICT; |
| 1247 | |
| 1248 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; |
| 1249 | end = start + len; |
| 1250 | |
| 1251 | if (end < start) |
| 1252 | return -EINVAL; |
| 1253 | if (end == start) |
| 1254 | return 0; |
| 1255 | |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1256 | new = mpol_new(mode, mode_flags, nmask); |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1257 | if (IS_ERR(new)) |
| 1258 | return PTR_ERR(new); |
| 1259 | |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 1260 | if (flags & MPOL_MF_LAZY) |
| 1261 | new->flags |= MPOL_F_MOF; |
| 1262 | |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1263 | /* |
| 1264 | * If we are using the default policy then operation |
| 1265 | * on discontinuous address spaces is okay after all |
| 1266 | */ |
| 1267 | if (!new) |
| 1268 | flags |= MPOL_MF_DISCONTIG_OK; |
| 1269 | |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1270 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
| 1271 | start, start + len, mode, mode_flags, |
David Rientjes | 00ef2d2 | 2013-02-22 16:35:36 -0800 | [diff] [blame] | 1272 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1273 | |
Christoph Lameter | 0aedadf | 2008-11-06 12:53:30 -0800 | [diff] [blame] | 1274 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
| 1275 | |
| 1276 | err = migrate_prep(); |
| 1277 | if (err) |
KOSAKI Motohiro | b05ca73 | 2009-10-26 16:49:59 -0700 | [diff] [blame] | 1278 | goto mpol_out; |
Christoph Lameter | 0aedadf | 2008-11-06 12:53:30 -0800 | [diff] [blame] | 1279 | } |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 1280 | { |
| 1281 | NODEMASK_SCRATCH(scratch); |
| 1282 | if (scratch) { |
| 1283 | down_write(&mm->mmap_sem); |
| 1284 | task_lock(current); |
| 1285 | err = mpol_set_nodemask(new, nmask, scratch); |
| 1286 | task_unlock(current); |
| 1287 | if (err) |
| 1288 | up_write(&mm->mmap_sem); |
| 1289 | } else |
| 1290 | err = -ENOMEM; |
| 1291 | NODEMASK_SCRATCH_FREE(scratch); |
| 1292 | } |
KOSAKI Motohiro | b05ca73 | 2009-10-26 16:49:59 -0700 | [diff] [blame] | 1293 | if (err) |
| 1294 | goto mpol_out; |
| 1295 | |
Naoya Horiguchi | 9809494 | 2013-09-11 14:22:14 -0700 | [diff] [blame^] | 1296 | vma = queue_pages_range(mm, start, end, nmask, |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1297 | flags | MPOL_MF_INVERT, &pagelist); |
| 1298 | |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 1299 | err = PTR_ERR(vma); /* maybe ... */ |
Mel Gorman | a720094 | 2012-11-16 09:37:58 +0000 | [diff] [blame] | 1300 | if (!IS_ERR(vma)) |
KOSAKI Motohiro | 9d8cebd | 2010-03-05 13:41:57 -0800 | [diff] [blame] | 1301 | err = mbind_range(mm, start, end, new); |
Christoph Lameter | 7e2ab15 | 2006-02-01 03:05:40 -0800 | [diff] [blame] | 1302 | |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 1303 | if (!err) { |
| 1304 | int nr_failed = 0; |
| 1305 | |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1306 | if (!list_empty(&pagelist)) { |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 1307 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
Christoph Lameter | 95a402c | 2006-06-23 02:03:53 -0700 | [diff] [blame] | 1308 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
Hugh Dickins | 9c620e2 | 2013-02-22 16:35:14 -0800 | [diff] [blame] | 1309 | (unsigned long)vma, |
| 1310 | MIGRATE_SYNC, MR_MEMPOLICY_MBIND); |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1311 | if (nr_failed) |
Naoya Horiguchi | 74060e4 | 2013-09-11 14:22:06 -0700 | [diff] [blame] | 1312 | putback_movable_pages(&pagelist); |
Minchan Kim | cf608ac | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 1313 | } |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1314 | |
Lee Schermerhorn | b24f53a | 2012-10-25 14:16:32 +0200 | [diff] [blame] | 1315 | if (nr_failed && (flags & MPOL_MF_STRICT)) |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1316 | err = -EIO; |
KOSAKI Motohiro | ab8a3e1 | 2009-10-26 16:49:58 -0700 | [diff] [blame] | 1317 | } else |
| 1318 | putback_lru_pages(&pagelist); |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 1319 | |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1320 | up_write(&mm->mmap_sem); |
KOSAKI Motohiro | b05ca73 | 2009-10-26 16:49:59 -0700 | [diff] [blame] | 1321 | mpol_out: |
Lee Schermerhorn | f0be3d3 | 2008-04-28 02:13:08 -0700 | [diff] [blame] | 1322 | mpol_put(new); |
Christoph Lameter | 6ce3c4c | 2006-01-08 01:01:04 -0800 | [diff] [blame] | 1323 | return err; |
| 1324 | } |
| 1325 | |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1326 | /* |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1327 | * User space interface with variable sized bitmaps for nodelists. |
| 1328 | */ |
| 1329 | |
| 1330 | /* Copy a node mask from user space. */ |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1331 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1332 | unsigned long maxnode) |
| 1333 | { |
| 1334 | unsigned long k; |
| 1335 | unsigned long nlongs; |
| 1336 | unsigned long endmask; |
| 1337 | |
| 1338 | --maxnode; |
| 1339 | nodes_clear(*nodes); |
| 1340 | if (maxnode == 0 || !nmask) |
| 1341 | return 0; |
Andi Kleen | a9c930b | 2006-02-20 18:27:59 -0800 | [diff] [blame] | 1342 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
Chris Wright | 636f13c | 2006-02-17 13:59:36 -0800 | [diff] [blame] | 1343 | return -EINVAL; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1344 | |
| 1345 | nlongs = BITS_TO_LONGS(maxnode); |
| 1346 | if ((maxnode % BITS_PER_LONG) == 0) |
| 1347 | endmask = ~0UL; |
| 1348 | else |
| 1349 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; |
| 1350 | |
| 1351 | /* When the user specified more nodes than supported just check |
| 1352 | if the non supported part is all zero. */ |
| 1353 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { |
| 1354 | if (nlongs > PAGE_SIZE/sizeof(long)) |
| 1355 | return -EINVAL; |
| 1356 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { |
| 1357 | unsigned long t; |
| 1358 | if (get_user(t, nmask + k)) |
| 1359 | return -EFAULT; |
| 1360 | if (k == nlongs - 1) { |
| 1361 | if (t & endmask) |
| 1362 | return -EINVAL; |
| 1363 | } else if (t) |
| 1364 | return -EINVAL; |
| 1365 | } |
| 1366 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); |
| 1367 | endmask = ~0UL; |
| 1368 | } |
| 1369 | |
| 1370 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) |
| 1371 | return -EFAULT; |
| 1372 | nodes_addr(*nodes)[nlongs-1] &= endmask; |
| 1373 | return 0; |
| 1374 | } |
| 1375 | |
| 1376 | /* Copy a kernel node mask to user space */ |
| 1377 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, |
| 1378 | nodemask_t *nodes) |
| 1379 | { |
| 1380 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; |
| 1381 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); |
| 1382 | |
| 1383 | if (copy > nbytes) { |
| 1384 | if (copy > PAGE_SIZE) |
| 1385 | return -EINVAL; |
| 1386 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) |
| 1387 | return -EFAULT; |
| 1388 | copy = nbytes; |
| 1389 | } |
| 1390 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; |
| 1391 | } |
| 1392 | |
Heiko Carstens | 938bb9f | 2009-01-14 14:14:30 +0100 | [diff] [blame] | 1393 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
| 1394 | unsigned long, mode, unsigned long __user *, nmask, |
| 1395 | unsigned long, maxnode, unsigned, flags) |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1396 | { |
| 1397 | nodemask_t nodes; |
| 1398 | int err; |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1399 | unsigned short mode_flags; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1400 | |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1401 | mode_flags = mode & MPOL_MODE_FLAGS; |
| 1402 | mode &= ~MPOL_MODE_FLAGS; |
David Rientjes | a3b51e0 | 2008-04-28 02:12:23 -0700 | [diff] [blame] | 1403 | if (mode >= MPOL_MAX) |
| 1404 | return -EINVAL; |
David Rientjes | 4c50bc0 | 2008-04-28 02:12:30 -0700 | [diff] [blame] | 1405 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
| 1406 | (mode_flags & MPOL_F_RELATIVE_NODES)) |
| 1407 | return -EINVAL; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1408 | err = get_nodes(&nodes, nmask, maxnode); |
| 1409 | if (err) |
| 1410 | return err; |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1411 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1412 | } |
| 1413 | |
| 1414 | /* Set the process memory policy */ |
Heiko Carstens | 938bb9f | 2009-01-14 14:14:30 +0100 | [diff] [blame] | 1415 | SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask, |
| 1416 | unsigned long, maxnode) |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1417 | { |
| 1418 | int err; |
| 1419 | nodemask_t nodes; |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1420 | unsigned short flags; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1421 | |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1422 | flags = mode & MPOL_MODE_FLAGS; |
| 1423 | mode &= ~MPOL_MODE_FLAGS; |
| 1424 | if ((unsigned int)mode >= MPOL_MAX) |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1425 | return -EINVAL; |
David Rientjes | 4c50bc0 | 2008-04-28 02:12:30 -0700 | [diff] [blame] | 1426 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
| 1427 | return -EINVAL; |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1428 | err = get_nodes(&nodes, nmask, maxnode); |
| 1429 | if (err) |
| 1430 | return err; |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 1431 | return do_set_mempolicy(mode, flags, &nodes); |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1432 | } |
| 1433 | |
Heiko Carstens | 938bb9f | 2009-01-14 14:14:30 +0100 | [diff] [blame] | 1434 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
| 1435 | const unsigned long __user *, old_nodes, |
| 1436 | const unsigned long __user *, new_nodes) |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1437 | { |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1438 | const struct cred *cred = current_cred(), *tcred; |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1439 | struct mm_struct *mm = NULL; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1440 | struct task_struct *task; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1441 | nodemask_t task_nodes; |
| 1442 | int err; |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1443 | nodemask_t *old; |
| 1444 | nodemask_t *new; |
| 1445 | NODEMASK_SCRATCH(scratch); |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1446 | |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1447 | if (!scratch) |
| 1448 | return -ENOMEM; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1449 | |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1450 | old = &scratch->mask1; |
| 1451 | new = &scratch->mask2; |
| 1452 | |
| 1453 | err = get_nodes(old, old_nodes, maxnode); |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1454 | if (err) |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1455 | goto out; |
| 1456 | |
| 1457 | err = get_nodes(new, new_nodes, maxnode); |
| 1458 | if (err) |
| 1459 | goto out; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1460 | |
| 1461 | /* Find the mm_struct */ |
Zeng Zhaoming | 55cfaa3 | 2010-12-02 14:31:13 -0800 | [diff] [blame] | 1462 | rcu_read_lock(); |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 1463 | task = pid ? find_task_by_vpid(pid) : current; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1464 | if (!task) { |
Zeng Zhaoming | 55cfaa3 | 2010-12-02 14:31:13 -0800 | [diff] [blame] | 1465 | rcu_read_unlock(); |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1466 | err = -ESRCH; |
| 1467 | goto out; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1468 | } |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1469 | get_task_struct(task); |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1470 | |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1471 | err = -EINVAL; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1472 | |
| 1473 | /* |
| 1474 | * Check if this process has the right to modify the specified |
| 1475 | * process. The right exists if the process has administrative |
Alexey Dobriyan | 7f927fc | 2006-03-28 01:56:53 -0800 | [diff] [blame] | 1476 | * capabilities, superuser privileges or the same |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1477 | * userid as the target process. |
| 1478 | */ |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1479 | tcred = __task_cred(task); |
Eric W. Biederman | b38a86e | 2012-03-12 15:48:24 -0700 | [diff] [blame] | 1480 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
| 1481 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && |
Christoph Lameter | 74c0024 | 2006-03-14 19:50:21 -0800 | [diff] [blame] | 1482 | !capable(CAP_SYS_NICE)) { |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1483 | rcu_read_unlock(); |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1484 | err = -EPERM; |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1485 | goto out_put; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1486 | } |
David Howells | c69e8d9 | 2008-11-14 10:39:19 +1100 | [diff] [blame] | 1487 | rcu_read_unlock(); |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1488 | |
| 1489 | task_nodes = cpuset_mems_allowed(task); |
| 1490 | /* Is the user allowed to access the target nodes? */ |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1491 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1492 | err = -EPERM; |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1493 | goto out_put; |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1494 | } |
| 1495 | |
Lai Jiangshan | 01f13bd | 2012-12-12 13:51:33 -0800 | [diff] [blame] | 1496 | if (!nodes_subset(*new, node_states[N_MEMORY])) { |
Christoph Lameter | 3b42d28 | 2007-08-31 00:12:08 -0700 | [diff] [blame] | 1497 | err = -EINVAL; |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1498 | goto out_put; |
Christoph Lameter | 3b42d28 | 2007-08-31 00:12:08 -0700 | [diff] [blame] | 1499 | } |
| 1500 | |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 1501 | err = security_task_movememory(task); |
| 1502 | if (err) |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1503 | goto out_put; |
David Quigley | 86c3a76 | 2006-06-23 02:04:02 -0700 | [diff] [blame] | 1504 | |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1505 | mm = get_task_mm(task); |
| 1506 | put_task_struct(task); |
Sasha Levin | f2a9ef8 | 2012-04-25 16:01:52 -0700 | [diff] [blame] | 1507 | |
| 1508 | if (!mm) { |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1509 | err = -EINVAL; |
Sasha Levin | f2a9ef8 | 2012-04-25 16:01:52 -0700 | [diff] [blame] | 1510 | goto out; |
| 1511 | } |
| 1512 | |
| 1513 | err = do_migrate_pages(mm, old, new, |
| 1514 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1515 | |
| 1516 | mmput(mm); |
| 1517 | out: |
KOSAKI Motohiro | 596d7cf | 2010-08-09 17:19:01 -0700 | [diff] [blame] | 1518 | NODEMASK_SCRATCH_FREE(scratch); |
| 1519 | |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1520 | return err; |
Christoph Lameter | 3268c63 | 2012-03-21 16:34:06 -0700 | [diff] [blame] | 1521 | |
| 1522 | out_put: |
| 1523 | put_task_struct(task); |
| 1524 | goto out; |
| 1525 | |
Christoph Lameter | 3974388 | 2006-01-08 01:00:51 -0800 | [diff] [blame] | 1526 | } |
| 1527 | |
| 1528 | |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1529 | /* Retrieve NUMA policy */ |
Heiko Carstens | 938bb9f | 2009-01-14 14:14:30 +0100 | [diff] [blame] | 1530 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
| 1531 | unsigned long __user *, nmask, unsigned long, maxnode, |
| 1532 | unsigned long, addr, unsigned long, flags) |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1533 | { |
Adrian Bunk | dbcb0f1 | 2007-10-16 01:26:26 -0700 | [diff] [blame] | 1534 | int err; |
| 1535 | int uninitialized_var(pval); |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 1536 | nodemask_t nodes; |
| 1537 | |
| 1538 | if (nmask != NULL && maxnode < MAX_NUMNODES) |
| 1539 | return -EINVAL; |
| 1540 | |
| 1541 | err = do_get_mempolicy(&pval, &nodes, addr, flags); |
| 1542 | |
| 1543 | if (err) |
| 1544 | return err; |
| 1545 | |
| 1546 | if (policy && put_user(pval, policy)) |
| 1547 | return -EFAULT; |
| 1548 | |
| 1549 | if (nmask) |
| 1550 | err = copy_nodes_to_user(nmask, maxnode, &nodes); |
| 1551 | |
| 1552 | return err; |
| 1553 | } |
| 1554 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1555 | #ifdef CONFIG_COMPAT |
| 1556 | |
| 1557 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, |
| 1558 | compat_ulong_t __user *nmask, |
| 1559 | compat_ulong_t maxnode, |
| 1560 | compat_ulong_t addr, compat_ulong_t flags) |
| 1561 | { |
| 1562 | long err; |
| 1563 | unsigned long __user *nm = NULL; |
| 1564 | unsigned long nr_bits, alloc_size; |
| 1565 | DECLARE_BITMAP(bm, MAX_NUMNODES); |
| 1566 | |
| 1567 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); |
| 1568 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
| 1569 | |
| 1570 | if (nmask) |
| 1571 | nm = compat_alloc_user_space(alloc_size); |
| 1572 | |
| 1573 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); |
| 1574 | |
| 1575 | if (!err && nmask) { |
KAMEZAWA Hiroyuki | 2bbff6c | 2011-09-14 16:21:02 -0700 | [diff] [blame] | 1576 | unsigned long copy_size; |
| 1577 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); |
| 1578 | err = copy_from_user(bm, nm, copy_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1579 | /* ensure entire bitmap is zeroed */ |
| 1580 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); |
| 1581 | err |= compat_put_bitmap(nmask, bm, nr_bits); |
| 1582 | } |
| 1583 | |
| 1584 | return err; |
| 1585 | } |
| 1586 | |
| 1587 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, |
| 1588 | compat_ulong_t maxnode) |
| 1589 | { |
| 1590 | long err = 0; |
| 1591 | unsigned long __user *nm = NULL; |
| 1592 | unsigned long nr_bits, alloc_size; |
| 1593 | DECLARE_BITMAP(bm, MAX_NUMNODES); |
| 1594 | |
| 1595 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); |
| 1596 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
| 1597 | |
| 1598 | if (nmask) { |
| 1599 | err = compat_get_bitmap(bm, nmask, nr_bits); |
| 1600 | nm = compat_alloc_user_space(alloc_size); |
| 1601 | err |= copy_to_user(nm, bm, alloc_size); |
| 1602 | } |
| 1603 | |
| 1604 | if (err) |
| 1605 | return -EFAULT; |
| 1606 | |
| 1607 | return sys_set_mempolicy(mode, nm, nr_bits+1); |
| 1608 | } |
| 1609 | |
| 1610 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, |
| 1611 | compat_ulong_t mode, compat_ulong_t __user *nmask, |
| 1612 | compat_ulong_t maxnode, compat_ulong_t flags) |
| 1613 | { |
| 1614 | long err = 0; |
| 1615 | unsigned long __user *nm = NULL; |
| 1616 | unsigned long nr_bits, alloc_size; |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1617 | nodemask_t bm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1618 | |
| 1619 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); |
| 1620 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
| 1621 | |
| 1622 | if (nmask) { |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1623 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1624 | nm = compat_alloc_user_space(alloc_size); |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1625 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | } |
| 1627 | |
| 1628 | if (err) |
| 1629 | return -EFAULT; |
| 1630 | |
| 1631 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); |
| 1632 | } |
| 1633 | |
| 1634 | #endif |
| 1635 | |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 1636 | /* |
| 1637 | * get_vma_policy(@task, @vma, @addr) |
| 1638 | * @task - task for fallback if vma policy == default |
| 1639 | * @vma - virtual memory area whose policy is sought |
| 1640 | * @addr - address in @vma for shared policy lookup |
| 1641 | * |
| 1642 | * Returns effective policy for a VMA at specified address. |
| 1643 | * Falls back to @task or system default policy, as necessary. |
David Rientjes | 32f8516 | 2012-10-16 17:31:23 -0700 | [diff] [blame] | 1644 | * Current or other task's task mempolicy and non-shared vma policies must be |
| 1645 | * protected by task_lock(task) by the caller. |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1646 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
| 1647 | * count--added by the get_policy() vm_op, as appropriate--to protect against |
| 1648 | * freeing by another task. It is the caller's responsibility to free the |
| 1649 | * extra reference for shared policies. |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 1650 | */ |
Stephen Wilson | d98f6cb | 2011-05-24 17:12:41 -0700 | [diff] [blame] | 1651 | struct mempolicy *get_vma_policy(struct task_struct *task, |
Christoph Lameter | 48fce34 | 2006-01-08 01:01:03 -0800 | [diff] [blame] | 1652 | struct vm_area_struct *vma, unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1653 | { |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 1654 | struct mempolicy *pol = get_task_policy(task); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1655 | |
| 1656 | if (vma) { |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 1657 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
Lee Schermerhorn | ae4d8c1 | 2008-04-28 02:13:11 -0700 | [diff] [blame] | 1658 | struct mempolicy *vpol = vma->vm_ops->get_policy(vma, |
| 1659 | addr); |
| 1660 | if (vpol) |
| 1661 | pol = vpol; |
Mel Gorman | 00442ad | 2012-10-08 16:29:20 -0700 | [diff] [blame] | 1662 | } else if (vma->vm_policy) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1663 | pol = vma->vm_policy; |
Mel Gorman | 00442ad | 2012-10-08 16:29:20 -0700 | [diff] [blame] | 1664 | |
| 1665 | /* |
| 1666 | * shmem_alloc_page() passes MPOL_F_SHARED policy with |
| 1667 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference |
| 1668 | * count on these policies which will be dropped by |
| 1669 | * mpol_cond_put() later |
| 1670 | */ |
| 1671 | if (mpol_needs_cond_ref(pol)) |
| 1672 | mpol_get(pol); |
| 1673 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1674 | } |
| 1675 | if (!pol) |
| 1676 | pol = &default_policy; |
| 1677 | return pol; |
| 1678 | } |
| 1679 | |
Lai Jiangshan | d3eb157 | 2013-02-22 16:33:22 -0800 | [diff] [blame] | 1680 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
| 1681 | { |
| 1682 | enum zone_type dynamic_policy_zone = policy_zone; |
| 1683 | |
| 1684 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); |
| 1685 | |
| 1686 | /* |
| 1687 | * if policy->v.nodes has movable memory only, |
| 1688 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. |
| 1689 | * |
| 1690 | * policy->v.nodes is intersect with node_states[N_MEMORY]. |
| 1691 | * so if the following test faile, it implies |
| 1692 | * policy->v.nodes has movable memory only. |
| 1693 | */ |
| 1694 | if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY])) |
| 1695 | dynamic_policy_zone = ZONE_MOVABLE; |
| 1696 | |
| 1697 | return zone >= dynamic_policy_zone; |
| 1698 | } |
| 1699 | |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1700 | /* |
| 1701 | * Return a nodemask representing a mempolicy for filtering nodes for |
| 1702 | * page allocation |
| 1703 | */ |
| 1704 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1705 | { |
| 1706 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 1707 | if (unlikely(policy->mode == MPOL_BIND) && |
Lai Jiangshan | d3eb157 | 2013-02-22 16:33:22 -0800 | [diff] [blame] | 1708 | apply_policy_zone(policy, gfp_zone(gfp)) && |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1709 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) |
| 1710 | return &policy->v.nodes; |
| 1711 | |
| 1712 | return NULL; |
| 1713 | } |
| 1714 | |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1715 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
Andi Kleen | 2f5f948 | 2011-03-04 17:36:29 -0800 | [diff] [blame] | 1716 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy, |
| 1717 | int nd) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1718 | { |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 1719 | switch (policy->mode) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1720 | case MPOL_PREFERRED: |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 1721 | if (!(policy->flags & MPOL_F_LOCAL)) |
| 1722 | nd = policy->v.preferred_node; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1723 | break; |
| 1724 | case MPOL_BIND: |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1725 | /* |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1726 | * Normally, MPOL_BIND allocations are node-local within the |
| 1727 | * allowed nodemask. However, if __GFP_THISNODE is set and the |
Bob Liu | 6eb27e1 | 2010-05-24 14:32:00 -0700 | [diff] [blame] | 1728 | * current node isn't part of the mask, we use the zonelist for |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1729 | * the first node in the mask instead. |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1730 | */ |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1731 | if (unlikely(gfp & __GFP_THISNODE) && |
| 1732 | unlikely(!node_isset(nd, policy->v.nodes))) |
| 1733 | nd = first_node(policy->v.nodes); |
| 1734 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1735 | default: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1736 | BUG(); |
| 1737 | } |
Mel Gorman | 0e88460 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 1738 | return node_zonelist(nd, gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1739 | } |
| 1740 | |
| 1741 | /* Do dynamic interleaving for a process */ |
| 1742 | static unsigned interleave_nodes(struct mempolicy *policy) |
| 1743 | { |
| 1744 | unsigned nid, next; |
| 1745 | struct task_struct *me = current; |
| 1746 | |
| 1747 | nid = me->il_next; |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1748 | next = next_node(nid, policy->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1749 | if (next >= MAX_NUMNODES) |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1750 | next = first_node(policy->v.nodes); |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 1751 | if (next < MAX_NUMNODES) |
| 1752 | me->il_next = next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | return nid; |
| 1754 | } |
| 1755 | |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1756 | /* |
| 1757 | * Depending on the memory policy provide a node from which to allocate the |
| 1758 | * next slab entry. |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1759 | * @policy must be protected by freeing by the caller. If @policy is |
| 1760 | * the current task's mempolicy, this protection is implicit, as only the |
| 1761 | * task can change it's policy. The system default policy requires no |
| 1762 | * such protection. |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1763 | */ |
Andi Kleen | e7b691b | 2012-06-09 02:40:03 -0700 | [diff] [blame] | 1764 | unsigned slab_node(void) |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1765 | { |
Andi Kleen | e7b691b | 2012-06-09 02:40:03 -0700 | [diff] [blame] | 1766 | struct mempolicy *policy; |
| 1767 | |
| 1768 | if (in_interrupt()) |
| 1769 | return numa_node_id(); |
| 1770 | |
| 1771 | policy = current->mempolicy; |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 1772 | if (!policy || policy->flags & MPOL_F_LOCAL) |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 1773 | return numa_node_id(); |
Christoph Lameter | 765c450 | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 1774 | |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 1775 | switch (policy->mode) { |
| 1776 | case MPOL_PREFERRED: |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 1777 | /* |
| 1778 | * handled MPOL_F_LOCAL above |
| 1779 | */ |
| 1780 | return policy->v.preferred_node; |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 1781 | |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1782 | case MPOL_INTERLEAVE: |
| 1783 | return interleave_nodes(policy); |
| 1784 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1785 | case MPOL_BIND: { |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1786 | /* |
| 1787 | * Follow bind policy behavior and start allocation at the |
| 1788 | * first node. |
| 1789 | */ |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1790 | struct zonelist *zonelist; |
| 1791 | struct zone *zone; |
| 1792 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
| 1793 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; |
| 1794 | (void)first_zones_zonelist(zonelist, highest_zoneidx, |
| 1795 | &policy->v.nodes, |
| 1796 | &zone); |
Eric Dumazet | 800416f | 2010-10-27 19:33:43 +0200 | [diff] [blame] | 1797 | return zone ? zone->node : numa_node_id(); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1798 | } |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1799 | |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1800 | default: |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 1801 | BUG(); |
Christoph Lameter | dc85da1 | 2006-01-18 17:42:36 -0800 | [diff] [blame] | 1802 | } |
| 1803 | } |
| 1804 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1805 | /* Do static interleaving for a VMA with known offset. */ |
| 1806 | static unsigned offset_il_node(struct mempolicy *pol, |
| 1807 | struct vm_area_struct *vma, unsigned long off) |
| 1808 | { |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1809 | unsigned nnodes = nodes_weight(pol->v.nodes); |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 1810 | unsigned target; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1811 | int c; |
| 1812 | int nid = -1; |
| 1813 | |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 1814 | if (!nnodes) |
| 1815 | return numa_node_id(); |
| 1816 | target = (unsigned int)off % nnodes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1817 | c = 0; |
| 1818 | do { |
Andi Kleen | dfcd3c0 | 2005-10-29 18:15:48 -0700 | [diff] [blame] | 1819 | nid = next_node(nid, pol->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1820 | c++; |
| 1821 | } while (c <= target); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1822 | return nid; |
| 1823 | } |
| 1824 | |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1825 | /* Determine a node number for interleave */ |
| 1826 | static inline unsigned interleave_nid(struct mempolicy *pol, |
| 1827 | struct vm_area_struct *vma, unsigned long addr, int shift) |
| 1828 | { |
| 1829 | if (vma) { |
| 1830 | unsigned long off; |
| 1831 | |
Nishanth Aravamudan | 3b98b08 | 2006-08-31 21:27:53 -0700 | [diff] [blame] | 1832 | /* |
| 1833 | * for small pages, there is no difference between |
| 1834 | * shift and PAGE_SHIFT, so the bit-shift is safe. |
| 1835 | * for huge pages, since vm_pgoff is in units of small |
| 1836 | * pages, we need to shift off the always 0 bits to get |
| 1837 | * a useful offset. |
| 1838 | */ |
| 1839 | BUG_ON(shift < PAGE_SHIFT); |
| 1840 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1841 | off += (addr - vma->vm_start) >> shift; |
| 1842 | return offset_il_node(pol, vma, off); |
| 1843 | } else |
| 1844 | return interleave_nodes(pol); |
| 1845 | } |
| 1846 | |
Michal Hocko | 778d3b0 | 2011-07-26 16:08:30 -0700 | [diff] [blame] | 1847 | /* |
| 1848 | * Return the bit number of a random bit set in the nodemask. |
| 1849 | * (returns -1 if nodemask is empty) |
| 1850 | */ |
| 1851 | int node_random(const nodemask_t *maskp) |
| 1852 | { |
| 1853 | int w, bit = -1; |
| 1854 | |
| 1855 | w = nodes_weight(*maskp); |
| 1856 | if (w) |
| 1857 | bit = bitmap_ord_to_pos(maskp->bits, |
| 1858 | get_random_int() % w, MAX_NUMNODES); |
| 1859 | return bit; |
| 1860 | } |
| 1861 | |
Chen, Kenneth W | 00ac59a | 2006-02-03 21:51:14 +0100 | [diff] [blame] | 1862 | #ifdef CONFIG_HUGETLBFS |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 1863 | /* |
| 1864 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) |
| 1865 | * @vma = virtual memory area whose policy is sought |
| 1866 | * @addr = address in @vma for shared policy lookup and interleave policy |
| 1867 | * @gfp_flags = for requested zone |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1868 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
| 1869 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 1870 | * |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1871 | * Returns a zonelist suitable for a huge page allocation and a pointer |
| 1872 | * to the struct mempolicy for conditional unref after allocation. |
| 1873 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's |
| 1874 | * @nodemask for filtering the zonelist. |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 1875 | * |
| 1876 | * Must be protected by get_mems_allowed() |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 1877 | */ |
Mel Gorman | 396faf0 | 2007-07-17 04:03:13 -0700 | [diff] [blame] | 1878 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1879 | gfp_t gfp_flags, struct mempolicy **mpol, |
| 1880 | nodemask_t **nodemask) |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1881 | { |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 1882 | struct zonelist *zl; |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1883 | |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1884 | *mpol = get_vma_policy(current, vma, addr); |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1885 | *nodemask = NULL; /* assume !MPOL_BIND */ |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1886 | |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1887 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
| 1888 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 1889 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1890 | } else { |
Andi Kleen | 2f5f948 | 2011-03-04 17:36:29 -0800 | [diff] [blame] | 1891 | zl = policy_zonelist(gfp_flags, *mpol, numa_node_id()); |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 1892 | if ((*mpol)->mode == MPOL_BIND) |
| 1893 | *nodemask = &(*mpol)->v.nodes; |
Lee Schermerhorn | 480eccf | 2007-09-18 22:46:47 -0700 | [diff] [blame] | 1894 | } |
| 1895 | return zl; |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1896 | } |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1897 | |
| 1898 | /* |
| 1899 | * init_nodemask_of_mempolicy |
| 1900 | * |
| 1901 | * If the current task's mempolicy is "default" [NULL], return 'false' |
| 1902 | * to indicate default policy. Otherwise, extract the policy nodemask |
| 1903 | * for 'bind' or 'interleave' policy into the argument nodemask, or |
| 1904 | * initialize the argument nodemask to contain the single node for |
| 1905 | * 'preferred' or 'local' policy and return 'true' to indicate presence |
| 1906 | * of non-default mempolicy. |
| 1907 | * |
| 1908 | * We don't bother with reference counting the mempolicy [mpol_get/put] |
| 1909 | * because the current task is examining it's own mempolicy and a task's |
| 1910 | * mempolicy is only ever changed by the task itself. |
| 1911 | * |
| 1912 | * N.B., it is the caller's responsibility to free a returned nodemask. |
| 1913 | */ |
| 1914 | bool init_nodemask_of_mempolicy(nodemask_t *mask) |
| 1915 | { |
| 1916 | struct mempolicy *mempolicy; |
| 1917 | int nid; |
| 1918 | |
| 1919 | if (!(mask && current->mempolicy)) |
| 1920 | return false; |
| 1921 | |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 1922 | task_lock(current); |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1923 | mempolicy = current->mempolicy; |
| 1924 | switch (mempolicy->mode) { |
| 1925 | case MPOL_PREFERRED: |
| 1926 | if (mempolicy->flags & MPOL_F_LOCAL) |
| 1927 | nid = numa_node_id(); |
| 1928 | else |
| 1929 | nid = mempolicy->v.preferred_node; |
| 1930 | init_nodemask_of_node(mask, nid); |
| 1931 | break; |
| 1932 | |
| 1933 | case MPOL_BIND: |
| 1934 | /* Fall through */ |
| 1935 | case MPOL_INTERLEAVE: |
| 1936 | *mask = mempolicy->v.nodes; |
| 1937 | break; |
| 1938 | |
| 1939 | default: |
| 1940 | BUG(); |
| 1941 | } |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 1942 | task_unlock(current); |
Lee Schermerhorn | 06808b0 | 2009-12-14 17:58:21 -0800 | [diff] [blame] | 1943 | |
| 1944 | return true; |
| 1945 | } |
Chen, Kenneth W | 00ac59a | 2006-02-03 21:51:14 +0100 | [diff] [blame] | 1946 | #endif |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 1947 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 1948 | /* |
| 1949 | * mempolicy_nodemask_intersects |
| 1950 | * |
| 1951 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default |
| 1952 | * policy. Otherwise, check for intersection between mask and the policy |
| 1953 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' |
| 1954 | * policy, always return true since it may allocate elsewhere on fallback. |
| 1955 | * |
| 1956 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. |
| 1957 | */ |
| 1958 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, |
| 1959 | const nodemask_t *mask) |
| 1960 | { |
| 1961 | struct mempolicy *mempolicy; |
| 1962 | bool ret = true; |
| 1963 | |
| 1964 | if (!mask) |
| 1965 | return ret; |
| 1966 | task_lock(tsk); |
| 1967 | mempolicy = tsk->mempolicy; |
| 1968 | if (!mempolicy) |
| 1969 | goto out; |
| 1970 | |
| 1971 | switch (mempolicy->mode) { |
| 1972 | case MPOL_PREFERRED: |
| 1973 | /* |
| 1974 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to |
| 1975 | * allocate from, they may fallback to other nodes when oom. |
| 1976 | * Thus, it's possible for tsk to have allocated memory from |
| 1977 | * nodes in mask. |
| 1978 | */ |
| 1979 | break; |
| 1980 | case MPOL_BIND: |
| 1981 | case MPOL_INTERLEAVE: |
| 1982 | ret = nodes_intersects(mempolicy->v.nodes, *mask); |
| 1983 | break; |
| 1984 | default: |
| 1985 | BUG(); |
| 1986 | } |
| 1987 | out: |
| 1988 | task_unlock(tsk); |
| 1989 | return ret; |
| 1990 | } |
| 1991 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1992 | /* Allocate a page in interleaved policy. |
| 1993 | Own path because it needs to do special accounting. */ |
Andi Kleen | 662f3a0 | 2005-10-29 18:15:49 -0700 | [diff] [blame] | 1994 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
| 1995 | unsigned nid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1996 | { |
| 1997 | struct zonelist *zl; |
| 1998 | struct page *page; |
| 1999 | |
Mel Gorman | 0e88460 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 2000 | zl = node_zonelist(nid, gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2001 | page = __alloc_pages(gfp, order, zl); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2002 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
Christoph Lameter | ca889e6 | 2006-06-30 01:55:44 -0700 | [diff] [blame] | 2003 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2004 | return page; |
| 2005 | } |
| 2006 | |
| 2007 | /** |
Andrea Arcangeli | 0bbbc0b | 2011-01-13 15:47:05 -0800 | [diff] [blame] | 2008 | * alloc_pages_vma - Allocate a page for a VMA. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2009 | * |
| 2010 | * @gfp: |
| 2011 | * %GFP_USER user allocation. |
| 2012 | * %GFP_KERNEL kernel allocations, |
| 2013 | * %GFP_HIGHMEM highmem/user allocations, |
| 2014 | * %GFP_FS allocation should not call back into a file system. |
| 2015 | * %GFP_ATOMIC don't sleep. |
| 2016 | * |
Andrea Arcangeli | 0bbbc0b | 2011-01-13 15:47:05 -0800 | [diff] [blame] | 2017 | * @order:Order of the GFP allocation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2018 | * @vma: Pointer to VMA or NULL if not available. |
| 2019 | * @addr: Virtual Address of the allocation. Must be inside the VMA. |
| 2020 | * |
| 2021 | * This function allocates a page from the kernel page pool and applies |
| 2022 | * a NUMA policy associated with the VMA or the current process. |
| 2023 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the |
| 2024 | * mm_struct of the VMA to prevent it from going away. Should be used for |
| 2025 | * all allocations for pages that will be mapped into |
| 2026 | * user space. Returns NULL when no page can be allocated. |
| 2027 | * |
| 2028 | * Should be called with the mm_sem of the vma hold. |
| 2029 | */ |
| 2030 | struct page * |
Andrea Arcangeli | 0bbbc0b | 2011-01-13 15:47:05 -0800 | [diff] [blame] | 2031 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
Andi Kleen | 2f5f948 | 2011-03-04 17:36:29 -0800 | [diff] [blame] | 2032 | unsigned long addr, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2033 | { |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 2034 | struct mempolicy *pol; |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2035 | struct page *page; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 2036 | unsigned int cpuset_mems_cookie; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2037 | |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 2038 | retry_cpuset: |
| 2039 | pol = get_vma_policy(current, vma, addr); |
| 2040 | cpuset_mems_cookie = get_mems_allowed(); |
| 2041 | |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 2042 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2043 | unsigned nid; |
Christoph Lameter | 5da7ca8 | 2006-01-06 00:10:46 -0800 | [diff] [blame] | 2044 | |
Andi Kleen | 8eac563 | 2011-02-25 14:44:28 -0800 | [diff] [blame] | 2045 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 2046 | mpol_cond_put(pol); |
Andrea Arcangeli | 0bbbc0b | 2011-01-13 15:47:05 -0800 | [diff] [blame] | 2047 | page = alloc_page_interleave(gfp, order, nid); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 2048 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
| 2049 | goto retry_cpuset; |
| 2050 | |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2051 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2052 | } |
David Rientjes | 212a0a6 | 2012-12-11 16:02:51 -0800 | [diff] [blame] | 2053 | page = __alloc_pages_nodemask(gfp, order, |
| 2054 | policy_zonelist(gfp, pol, node), |
Andrea Arcangeli | 0bbbc0b | 2011-01-13 15:47:05 -0800 | [diff] [blame] | 2055 | policy_nodemask(gfp, pol)); |
David Rientjes | 212a0a6 | 2012-12-11 16:02:51 -0800 | [diff] [blame] | 2056 | if (unlikely(mpol_needs_cond_ref(pol))) |
| 2057 | __mpol_put(pol); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 2058 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
| 2059 | goto retry_cpuset; |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2060 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2061 | } |
| 2062 | |
| 2063 | /** |
| 2064 | * alloc_pages_current - Allocate pages. |
| 2065 | * |
| 2066 | * @gfp: |
| 2067 | * %GFP_USER user allocation, |
| 2068 | * %GFP_KERNEL kernel allocation, |
| 2069 | * %GFP_HIGHMEM highmem allocation, |
| 2070 | * %GFP_FS don't call back into a file system. |
| 2071 | * %GFP_ATOMIC don't sleep. |
| 2072 | * @order: Power of two of allocation size in pages. 0 is a single page. |
| 2073 | * |
| 2074 | * Allocate a page from the kernel page pool. When not in |
| 2075 | * interrupt context and apply the current process NUMA policy. |
| 2076 | * Returns NULL when no page can be allocated. |
| 2077 | * |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 2078 | * Don't call cpuset_update_task_memory_state() unless |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2079 | * 1) it's ok to take cpuset_sem (can WAIT), and |
| 2080 | * 2) allocating for current task (not interrupt). |
| 2081 | */ |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 2082 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2083 | { |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 2084 | struct mempolicy *pol = get_task_policy(current); |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2085 | struct page *page; |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 2086 | unsigned int cpuset_mems_cookie; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2087 | |
Christoph Lameter | 9b819d2 | 2006-09-25 23:31:40 -0700 | [diff] [blame] | 2088 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2089 | pol = &default_policy; |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 2090 | |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 2091 | retry_cpuset: |
| 2092 | cpuset_mems_cookie = get_mems_allowed(); |
| 2093 | |
Lee Schermerhorn | 52cd3b0 | 2008-04-28 02:13:16 -0700 | [diff] [blame] | 2094 | /* |
| 2095 | * No reference counting needed for current->mempolicy |
| 2096 | * nor system default_policy |
| 2097 | */ |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 2098 | if (pol->mode == MPOL_INTERLEAVE) |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2099 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
| 2100 | else |
| 2101 | page = __alloc_pages_nodemask(gfp, order, |
Andi Kleen | 5c4b4be | 2011-03-04 17:36:32 -0800 | [diff] [blame] | 2102 | policy_zonelist(gfp, pol, numa_node_id()), |
| 2103 | policy_nodemask(gfp, pol)); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 2104 | |
| 2105 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
| 2106 | goto retry_cpuset; |
| 2107 | |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2108 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2109 | } |
| 2110 | EXPORT_SYMBOL(alloc_pages_current); |
| 2111 | |
Oleg Nesterov | ef0855d | 2013-09-11 14:20:14 -0700 | [diff] [blame] | 2112 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
| 2113 | { |
| 2114 | struct mempolicy *pol = mpol_dup(vma_policy(src)); |
| 2115 | |
| 2116 | if (IS_ERR(pol)) |
| 2117 | return PTR_ERR(pol); |
| 2118 | dst->vm_policy = pol; |
| 2119 | return 0; |
| 2120 | } |
| 2121 | |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 2122 | /* |
Lee Schermerhorn | 846a16b | 2008-04-28 02:13:09 -0700 | [diff] [blame] | 2123 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 2124 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
| 2125 | * with the mems_allowed returned by cpuset_mems_allowed(). This |
| 2126 | * keeps mempolicies cpuset relative after its cpuset moves. See |
| 2127 | * further kernel/cpuset.c update_nodemask(). |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 2128 | * |
| 2129 | * current's mempolicy may be rebinded by the other task(the task that changes |
| 2130 | * cpuset's mems), so we needn't do rebind work for current task. |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 2131 | */ |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 2132 | |
Lee Schermerhorn | 846a16b | 2008-04-28 02:13:09 -0700 | [diff] [blame] | 2133 | /* Slow path of a mempolicy duplicate */ |
| 2134 | struct mempolicy *__mpol_dup(struct mempolicy *old) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2135 | { |
| 2136 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
| 2137 | |
| 2138 | if (!new) |
| 2139 | return ERR_PTR(-ENOMEM); |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 2140 | |
| 2141 | /* task's mempolicy is protected by alloc_lock */ |
| 2142 | if (old == current->mempolicy) { |
| 2143 | task_lock(current); |
| 2144 | *new = *old; |
| 2145 | task_unlock(current); |
| 2146 | } else |
| 2147 | *new = *old; |
| 2148 | |
Paul E. McKenney | 99ee4ca | 2010-03-03 17:50:17 -0800 | [diff] [blame] | 2149 | rcu_read_lock(); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 2150 | if (current_cpuset_is_being_rebound()) { |
| 2151 | nodemask_t mems = cpuset_mems_allowed(current); |
Miao Xie | 708c1bb | 2010-05-24 14:32:07 -0700 | [diff] [blame] | 2152 | if (new->flags & MPOL_F_REBINDING) |
| 2153 | mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2); |
| 2154 | else |
| 2155 | mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 2156 | } |
Paul E. McKenney | 99ee4ca | 2010-03-03 17:50:17 -0800 | [diff] [blame] | 2157 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2158 | atomic_set(&new->refcnt, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2159 | return new; |
| 2160 | } |
| 2161 | |
| 2162 | /* Slow path of a mempolicy comparison */ |
KOSAKI Motohiro | fcfb4dc | 2012-01-10 15:08:21 -0800 | [diff] [blame] | 2163 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2164 | { |
| 2165 | if (!a || !b) |
KOSAKI Motohiro | fcfb4dc | 2012-01-10 15:08:21 -0800 | [diff] [blame] | 2166 | return false; |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 2167 | if (a->mode != b->mode) |
KOSAKI Motohiro | fcfb4dc | 2012-01-10 15:08:21 -0800 | [diff] [blame] | 2168 | return false; |
Bob Liu | 1980050 | 2010-05-24 14:32:01 -0700 | [diff] [blame] | 2169 | if (a->flags != b->flags) |
KOSAKI Motohiro | fcfb4dc | 2012-01-10 15:08:21 -0800 | [diff] [blame] | 2170 | return false; |
Bob Liu | 1980050 | 2010-05-24 14:32:01 -0700 | [diff] [blame] | 2171 | if (mpol_store_user_nodemask(a)) |
| 2172 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) |
KOSAKI Motohiro | fcfb4dc | 2012-01-10 15:08:21 -0800 | [diff] [blame] | 2173 | return false; |
Bob Liu | 1980050 | 2010-05-24 14:32:01 -0700 | [diff] [blame] | 2174 | |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 2175 | switch (a->mode) { |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2176 | case MPOL_BIND: |
| 2177 | /* Fall through */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2178 | case MPOL_INTERLEAVE: |
KOSAKI Motohiro | fcfb4dc | 2012-01-10 15:08:21 -0800 | [diff] [blame] | 2179 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2180 | case MPOL_PREFERRED: |
Namhyung Kim | 7571966 | 2011-03-22 16:33:02 -0700 | [diff] [blame] | 2181 | return a->v.preferred_node == b->v.preferred_node; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2182 | default: |
| 2183 | BUG(); |
KOSAKI Motohiro | fcfb4dc | 2012-01-10 15:08:21 -0800 | [diff] [blame] | 2184 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2185 | } |
| 2186 | } |
| 2187 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2188 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2189 | * Shared memory backing store policy support. |
| 2190 | * |
| 2191 | * Remember policies even when nobody has shared memory mapped. |
| 2192 | * The policies are kept in Red-Black tree linked from the inode. |
| 2193 | * They are protected by the sp->lock spinlock, which should be held |
| 2194 | * for any accesses to the tree. |
| 2195 | */ |
| 2196 | |
| 2197 | /* lookup first element intersecting start-end */ |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2198 | /* Caller holds sp->lock */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2199 | static struct sp_node * |
| 2200 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) |
| 2201 | { |
| 2202 | struct rb_node *n = sp->root.rb_node; |
| 2203 | |
| 2204 | while (n) { |
| 2205 | struct sp_node *p = rb_entry(n, struct sp_node, nd); |
| 2206 | |
| 2207 | if (start >= p->end) |
| 2208 | n = n->rb_right; |
| 2209 | else if (end <= p->start) |
| 2210 | n = n->rb_left; |
| 2211 | else |
| 2212 | break; |
| 2213 | } |
| 2214 | if (!n) |
| 2215 | return NULL; |
| 2216 | for (;;) { |
| 2217 | struct sp_node *w = NULL; |
| 2218 | struct rb_node *prev = rb_prev(n); |
| 2219 | if (!prev) |
| 2220 | break; |
| 2221 | w = rb_entry(prev, struct sp_node, nd); |
| 2222 | if (w->end <= start) |
| 2223 | break; |
| 2224 | n = prev; |
| 2225 | } |
| 2226 | return rb_entry(n, struct sp_node, nd); |
| 2227 | } |
| 2228 | |
| 2229 | /* Insert a new shared policy into the list. */ |
| 2230 | /* Caller holds sp->lock */ |
| 2231 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
| 2232 | { |
| 2233 | struct rb_node **p = &sp->root.rb_node; |
| 2234 | struct rb_node *parent = NULL; |
| 2235 | struct sp_node *nd; |
| 2236 | |
| 2237 | while (*p) { |
| 2238 | parent = *p; |
| 2239 | nd = rb_entry(parent, struct sp_node, nd); |
| 2240 | if (new->start < nd->start) |
| 2241 | p = &(*p)->rb_left; |
| 2242 | else if (new->end > nd->end) |
| 2243 | p = &(*p)->rb_right; |
| 2244 | else |
| 2245 | BUG(); |
| 2246 | } |
| 2247 | rb_link_node(&new->nd, parent, p); |
| 2248 | rb_insert_color(&new->nd, &sp->root); |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 2249 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 2250 | new->policy ? new->policy->mode : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2251 | } |
| 2252 | |
| 2253 | /* Find shared policy intersecting idx */ |
| 2254 | struct mempolicy * |
| 2255 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) |
| 2256 | { |
| 2257 | struct mempolicy *pol = NULL; |
| 2258 | struct sp_node *sn; |
| 2259 | |
| 2260 | if (!sp->root.rb_node) |
| 2261 | return NULL; |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2262 | spin_lock(&sp->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2263 | sn = sp_lookup(sp, idx, idx+1); |
| 2264 | if (sn) { |
| 2265 | mpol_get(sn->policy); |
| 2266 | pol = sn->policy; |
| 2267 | } |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2268 | spin_unlock(&sp->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2269 | return pol; |
| 2270 | } |
| 2271 | |
KOSAKI Motohiro | 63f74ca | 2012-10-08 16:29:19 -0700 | [diff] [blame] | 2272 | static void sp_free(struct sp_node *n) |
| 2273 | { |
| 2274 | mpol_put(n->policy); |
| 2275 | kmem_cache_free(sn_cache, n); |
| 2276 | } |
| 2277 | |
Lee Schermerhorn | 771fb4d | 2012-10-25 14:16:30 +0200 | [diff] [blame] | 2278 | /** |
| 2279 | * mpol_misplaced - check whether current page node is valid in policy |
| 2280 | * |
| 2281 | * @page - page to be checked |
| 2282 | * @vma - vm area where page mapped |
| 2283 | * @addr - virtual address where page mapped |
| 2284 | * |
| 2285 | * Lookup current policy node id for vma,addr and "compare to" page's |
| 2286 | * node id. |
| 2287 | * |
| 2288 | * Returns: |
| 2289 | * -1 - not misplaced, page is in the right node |
| 2290 | * node - node id where the page should be |
| 2291 | * |
| 2292 | * Policy determination "mimics" alloc_page_vma(). |
| 2293 | * Called from fault path where we know the vma and faulting address. |
| 2294 | */ |
| 2295 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) |
| 2296 | { |
| 2297 | struct mempolicy *pol; |
| 2298 | struct zone *zone; |
| 2299 | int curnid = page_to_nid(page); |
| 2300 | unsigned long pgoff; |
| 2301 | int polnid = -1; |
| 2302 | int ret = -1; |
| 2303 | |
| 2304 | BUG_ON(!vma); |
| 2305 | |
| 2306 | pol = get_vma_policy(current, vma, addr); |
| 2307 | if (!(pol->flags & MPOL_F_MOF)) |
| 2308 | goto out; |
| 2309 | |
| 2310 | switch (pol->mode) { |
| 2311 | case MPOL_INTERLEAVE: |
| 2312 | BUG_ON(addr >= vma->vm_end); |
| 2313 | BUG_ON(addr < vma->vm_start); |
| 2314 | |
| 2315 | pgoff = vma->vm_pgoff; |
| 2316 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; |
| 2317 | polnid = offset_il_node(pol, vma, pgoff); |
| 2318 | break; |
| 2319 | |
| 2320 | case MPOL_PREFERRED: |
| 2321 | if (pol->flags & MPOL_F_LOCAL) |
| 2322 | polnid = numa_node_id(); |
| 2323 | else |
| 2324 | polnid = pol->v.preferred_node; |
| 2325 | break; |
| 2326 | |
| 2327 | case MPOL_BIND: |
| 2328 | /* |
| 2329 | * allows binding to multiple nodes. |
| 2330 | * use current page if in policy nodemask, |
| 2331 | * else select nearest allowed node, if any. |
| 2332 | * If no allowed nodes, use current [!misplaced]. |
| 2333 | */ |
| 2334 | if (node_isset(curnid, pol->v.nodes)) |
| 2335 | goto out; |
| 2336 | (void)first_zones_zonelist( |
| 2337 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
| 2338 | gfp_zone(GFP_HIGHUSER), |
| 2339 | &pol->v.nodes, &zone); |
| 2340 | polnid = zone->node; |
| 2341 | break; |
| 2342 | |
| 2343 | default: |
| 2344 | BUG(); |
| 2345 | } |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 2346 | |
| 2347 | /* Migrate the page towards the node whose CPU is referencing it */ |
Mel Gorman | e42c8ff | 2012-11-12 09:17:07 +0000 | [diff] [blame] | 2348 | if (pol->flags & MPOL_F_MORON) { |
| 2349 | int last_nid; |
| 2350 | |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 2351 | polnid = numa_node_id(); |
| 2352 | |
Mel Gorman | e42c8ff | 2012-11-12 09:17:07 +0000 | [diff] [blame] | 2353 | /* |
| 2354 | * Multi-stage node selection is used in conjunction |
| 2355 | * with a periodic migration fault to build a temporal |
| 2356 | * task<->page relation. By using a two-stage filter we |
| 2357 | * remove short/unlikely relations. |
| 2358 | * |
| 2359 | * Using P(p) ~ n_p / n_t as per frequentist |
| 2360 | * probability, we can equate a task's usage of a |
| 2361 | * particular page (n_p) per total usage of this |
| 2362 | * page (n_t) (in a given time-span) to a probability. |
| 2363 | * |
| 2364 | * Our periodic faults will sample this probability and |
| 2365 | * getting the same result twice in a row, given these |
| 2366 | * samples are fully independent, is then given by |
| 2367 | * P(n)^2, provided our sample period is sufficiently |
| 2368 | * short compared to the usage pattern. |
| 2369 | * |
| 2370 | * This quadric squishes small probabilities, making |
| 2371 | * it less likely we act on an unlikely task<->page |
| 2372 | * relation. |
| 2373 | */ |
Mel Gorman | 22b751c | 2013-02-22 16:34:59 -0800 | [diff] [blame] | 2374 | last_nid = page_nid_xchg_last(page, polnid); |
Mel Gorman | e42c8ff | 2012-11-12 09:17:07 +0000 | [diff] [blame] | 2375 | if (last_nid != polnid) |
| 2376 | goto out; |
| 2377 | } |
| 2378 | |
Lee Schermerhorn | 771fb4d | 2012-10-25 14:16:30 +0200 | [diff] [blame] | 2379 | if (curnid != polnid) |
| 2380 | ret = polnid; |
| 2381 | out: |
| 2382 | mpol_cond_put(pol); |
| 2383 | |
| 2384 | return ret; |
| 2385 | } |
| 2386 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2387 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
| 2388 | { |
Paul Mundt | 140d5a4 | 2007-07-15 23:38:16 -0700 | [diff] [blame] | 2389 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2390 | rb_erase(&n->nd, &sp->root); |
KOSAKI Motohiro | 63f74ca | 2012-10-08 16:29:19 -0700 | [diff] [blame] | 2391 | sp_free(n); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2392 | } |
| 2393 | |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2394 | static void sp_node_init(struct sp_node *node, unsigned long start, |
| 2395 | unsigned long end, struct mempolicy *pol) |
| 2396 | { |
| 2397 | node->start = start; |
| 2398 | node->end = end; |
| 2399 | node->policy = pol; |
| 2400 | } |
| 2401 | |
Adrian Bunk | dbcb0f1 | 2007-10-16 01:26:26 -0700 | [diff] [blame] | 2402 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
| 2403 | struct mempolicy *pol) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | { |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 2405 | struct sp_node *n; |
| 2406 | struct mempolicy *newpol; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2407 | |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 2408 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2409 | if (!n) |
| 2410 | return NULL; |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 2411 | |
| 2412 | newpol = mpol_dup(pol); |
| 2413 | if (IS_ERR(newpol)) { |
| 2414 | kmem_cache_free(sn_cache, n); |
| 2415 | return NULL; |
| 2416 | } |
| 2417 | newpol->flags |= MPOL_F_SHARED; |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2418 | sp_node_init(n, start, end, newpol); |
KOSAKI Motohiro | 869833f | 2012-10-08 16:29:16 -0700 | [diff] [blame] | 2419 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2420 | return n; |
| 2421 | } |
| 2422 | |
| 2423 | /* Replace a policy range. */ |
| 2424 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, |
| 2425 | unsigned long end, struct sp_node *new) |
| 2426 | { |
Mel Gorman | b22d127 | 2012-10-08 16:29:17 -0700 | [diff] [blame] | 2427 | struct sp_node *n; |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2428 | struct sp_node *n_new = NULL; |
| 2429 | struct mempolicy *mpol_new = NULL; |
Mel Gorman | b22d127 | 2012-10-08 16:29:17 -0700 | [diff] [blame] | 2430 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2431 | |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2432 | restart: |
| 2433 | spin_lock(&sp->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2434 | n = sp_lookup(sp, start, end); |
| 2435 | /* Take care of old policies in the same range. */ |
| 2436 | while (n && n->start < end) { |
| 2437 | struct rb_node *next = rb_next(&n->nd); |
| 2438 | if (n->start >= start) { |
| 2439 | if (n->end <= end) |
| 2440 | sp_delete(sp, n); |
| 2441 | else |
| 2442 | n->start = end; |
| 2443 | } else { |
| 2444 | /* Old policy spanning whole new range. */ |
| 2445 | if (n->end > end) { |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2446 | if (!n_new) |
| 2447 | goto alloc_new; |
| 2448 | |
| 2449 | *mpol_new = *n->policy; |
| 2450 | atomic_set(&mpol_new->refcnt, 1); |
KOSAKI Motohiro | 7880639 | 2013-03-08 12:43:29 -0800 | [diff] [blame] | 2451 | sp_node_init(n_new, end, n->end, mpol_new); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2452 | n->end = start; |
Hillf Danton | 5ca3957 | 2013-03-08 12:43:28 -0800 | [diff] [blame] | 2453 | sp_insert(sp, n_new); |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2454 | n_new = NULL; |
| 2455 | mpol_new = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2456 | break; |
| 2457 | } else |
| 2458 | n->end = start; |
| 2459 | } |
| 2460 | if (!next) |
| 2461 | break; |
| 2462 | n = rb_entry(next, struct sp_node, nd); |
| 2463 | } |
| 2464 | if (new) |
| 2465 | sp_insert(sp, new); |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2466 | spin_unlock(&sp->lock); |
| 2467 | ret = 0; |
| 2468 | |
| 2469 | err_out: |
| 2470 | if (mpol_new) |
| 2471 | mpol_put(mpol_new); |
| 2472 | if (n_new) |
| 2473 | kmem_cache_free(sn_cache, n_new); |
| 2474 | |
Mel Gorman | b22d127 | 2012-10-08 16:29:17 -0700 | [diff] [blame] | 2475 | return ret; |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2476 | |
| 2477 | alloc_new: |
| 2478 | spin_unlock(&sp->lock); |
| 2479 | ret = -ENOMEM; |
| 2480 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
| 2481 | if (!n_new) |
| 2482 | goto err_out; |
| 2483 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
| 2484 | if (!mpol_new) |
| 2485 | goto err_out; |
| 2486 | goto restart; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2487 | } |
| 2488 | |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2489 | /** |
| 2490 | * mpol_shared_policy_init - initialize shared policy for inode |
| 2491 | * @sp: pointer to inode shared policy |
| 2492 | * @mpol: struct mempolicy to install |
| 2493 | * |
| 2494 | * Install non-NULL @mpol in inode's shared policy rb-tree. |
| 2495 | * On entry, the current task has a reference on a non-NULL @mpol. |
| 2496 | * This must be released on exit. |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 2497 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2498 | */ |
| 2499 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 2500 | { |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 2501 | int ret; |
| 2502 | |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2503 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2504 | spin_lock_init(&sp->lock); |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 2505 | |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2506 | if (mpol) { |
| 2507 | struct vm_area_struct pvma; |
| 2508 | struct mempolicy *new; |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 2509 | NODEMASK_SCRATCH(scratch); |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 2510 | |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 2511 | if (!scratch) |
Lee Schermerhorn | 5c0c165 | 2010-06-29 15:05:30 -0700 | [diff] [blame] | 2512 | goto put_mpol; |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2513 | /* contextualize the tmpfs mount point mempolicy */ |
| 2514 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); |
Lee Schermerhorn | 15d7783 | 2010-05-24 14:32:04 -0700 | [diff] [blame] | 2515 | if (IS_ERR(new)) |
Dan Carpenter | 0cae345 | 2010-05-25 23:42:58 -0700 | [diff] [blame] | 2516 | goto free_scratch; /* no valid nodemask intersection */ |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 2517 | |
| 2518 | task_lock(current); |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 2519 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
Miao Xie | 58568d2 | 2009-06-16 15:31:49 -0700 | [diff] [blame] | 2520 | task_unlock(current); |
Lee Schermerhorn | 15d7783 | 2010-05-24 14:32:04 -0700 | [diff] [blame] | 2521 | if (ret) |
Lee Schermerhorn | 5c0c165 | 2010-06-29 15:05:30 -0700 | [diff] [blame] | 2522 | goto put_new; |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 2523 | |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2524 | /* Create pseudo-vma that contains just the policy */ |
| 2525 | memset(&pvma, 0, sizeof(struct vm_area_struct)); |
| 2526 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ |
| 2527 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ |
Lee Schermerhorn | 15d7783 | 2010-05-24 14:32:04 -0700 | [diff] [blame] | 2528 | |
Lee Schermerhorn | 5c0c165 | 2010-06-29 15:05:30 -0700 | [diff] [blame] | 2529 | put_new: |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2530 | mpol_put(new); /* drop initial ref */ |
Dan Carpenter | 0cae345 | 2010-05-25 23:42:58 -0700 | [diff] [blame] | 2531 | free_scratch: |
KAMEZAWA Hiroyuki | 4bfc449 | 2009-08-06 15:07:33 -0700 | [diff] [blame] | 2532 | NODEMASK_SCRATCH_FREE(scratch); |
Lee Schermerhorn | 5c0c165 | 2010-06-29 15:05:30 -0700 | [diff] [blame] | 2533 | put_mpol: |
| 2534 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ |
Robin Holt | 7339ff8 | 2006-01-14 13:20:48 -0800 | [diff] [blame] | 2535 | } |
| 2536 | } |
| 2537 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2538 | int mpol_set_shared_policy(struct shared_policy *info, |
| 2539 | struct vm_area_struct *vma, struct mempolicy *npol) |
| 2540 | { |
| 2541 | int err; |
| 2542 | struct sp_node *new = NULL; |
| 2543 | unsigned long sz = vma_pages(vma); |
| 2544 | |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 2545 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2546 | vma->vm_pgoff, |
Lee Schermerhorn | 45c4745 | 2008-04-28 02:13:12 -0700 | [diff] [blame] | 2547 | sz, npol ? npol->mode : -1, |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 2548 | npol ? npol->flags : -1, |
David Rientjes | 00ef2d2 | 2013-02-22 16:35:36 -0800 | [diff] [blame] | 2549 | npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2550 | |
| 2551 | if (npol) { |
| 2552 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); |
| 2553 | if (!new) |
| 2554 | return -ENOMEM; |
| 2555 | } |
| 2556 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); |
| 2557 | if (err && new) |
KOSAKI Motohiro | 63f74ca | 2012-10-08 16:29:19 -0700 | [diff] [blame] | 2558 | sp_free(new); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2559 | return err; |
| 2560 | } |
| 2561 | |
| 2562 | /* Free a backing policy store on inode delete. */ |
| 2563 | void mpol_free_shared_policy(struct shared_policy *p) |
| 2564 | { |
| 2565 | struct sp_node *n; |
| 2566 | struct rb_node *next; |
| 2567 | |
| 2568 | if (!p->root.rb_node) |
| 2569 | return; |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2570 | spin_lock(&p->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2571 | next = rb_first(&p->root); |
| 2572 | while (next) { |
| 2573 | n = rb_entry(next, struct sp_node, nd); |
| 2574 | next = rb_next(&n->nd); |
KOSAKI Motohiro | 63f74ca | 2012-10-08 16:29:19 -0700 | [diff] [blame] | 2575 | sp_delete(p, n); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2576 | } |
Mel Gorman | 42288fe | 2012-12-21 23:10:25 +0000 | [diff] [blame] | 2577 | spin_unlock(&p->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2578 | } |
| 2579 | |
Mel Gorman | 1a687c2 | 2012-11-22 11:16:36 +0000 | [diff] [blame] | 2580 | #ifdef CONFIG_NUMA_BALANCING |
| 2581 | static bool __initdata numabalancing_override; |
| 2582 | |
| 2583 | static void __init check_numabalancing_enable(void) |
| 2584 | { |
| 2585 | bool numabalancing_default = false; |
| 2586 | |
| 2587 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) |
| 2588 | numabalancing_default = true; |
| 2589 | |
| 2590 | if (nr_node_ids > 1 && !numabalancing_override) { |
| 2591 | printk(KERN_INFO "Enabling automatic NUMA balancing. " |
| 2592 | "Configure with numa_balancing= or sysctl"); |
| 2593 | set_numabalancing_state(numabalancing_default); |
| 2594 | } |
| 2595 | } |
| 2596 | |
| 2597 | static int __init setup_numabalancing(char *str) |
| 2598 | { |
| 2599 | int ret = 0; |
| 2600 | if (!str) |
| 2601 | goto out; |
| 2602 | numabalancing_override = true; |
| 2603 | |
| 2604 | if (!strcmp(str, "enable")) { |
| 2605 | set_numabalancing_state(true); |
| 2606 | ret = 1; |
| 2607 | } else if (!strcmp(str, "disable")) { |
| 2608 | set_numabalancing_state(false); |
| 2609 | ret = 1; |
| 2610 | } |
| 2611 | out: |
| 2612 | if (!ret) |
| 2613 | printk(KERN_WARNING "Unable to parse numa_balancing=\n"); |
| 2614 | |
| 2615 | return ret; |
| 2616 | } |
| 2617 | __setup("numa_balancing=", setup_numabalancing); |
| 2618 | #else |
| 2619 | static inline void __init check_numabalancing_enable(void) |
| 2620 | { |
| 2621 | } |
| 2622 | #endif /* CONFIG_NUMA_BALANCING */ |
| 2623 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2624 | /* assumes fs == KERNEL_DS */ |
| 2625 | void __init numa_policy_init(void) |
| 2626 | { |
Paul Mundt | b71636e | 2007-07-15 23:38:15 -0700 | [diff] [blame] | 2627 | nodemask_t interleave_nodes; |
| 2628 | unsigned long largest = 0; |
| 2629 | int nid, prefer = 0; |
| 2630 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2631 | policy_cache = kmem_cache_create("numa_policy", |
| 2632 | sizeof(struct mempolicy), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 2633 | 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2634 | |
| 2635 | sn_cache = kmem_cache_create("shared_policy_node", |
| 2636 | sizeof(struct sp_node), |
Paul Mundt | 20c2df8 | 2007-07-20 10:11:58 +0900 | [diff] [blame] | 2637 | 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2638 | |
Mel Gorman | 5606e38 | 2012-11-02 18:19:13 +0000 | [diff] [blame] | 2639 | for_each_node(nid) { |
| 2640 | preferred_node_policy[nid] = (struct mempolicy) { |
| 2641 | .refcnt = ATOMIC_INIT(1), |
| 2642 | .mode = MPOL_PREFERRED, |
| 2643 | .flags = MPOL_F_MOF | MPOL_F_MORON, |
| 2644 | .v = { .preferred_node = nid, }, |
| 2645 | }; |
| 2646 | } |
| 2647 | |
Paul Mundt | b71636e | 2007-07-15 23:38:15 -0700 | [diff] [blame] | 2648 | /* |
| 2649 | * Set interleaving policy for system init. Interleaving is only |
| 2650 | * enabled across suitably sized nodes (default is >= 16MB), or |
| 2651 | * fall back to the largest node if they're all smaller. |
| 2652 | */ |
| 2653 | nodes_clear(interleave_nodes); |
Lai Jiangshan | 01f13bd | 2012-12-12 13:51:33 -0800 | [diff] [blame] | 2654 | for_each_node_state(nid, N_MEMORY) { |
Paul Mundt | b71636e | 2007-07-15 23:38:15 -0700 | [diff] [blame] | 2655 | unsigned long total_pages = node_present_pages(nid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2656 | |
Paul Mundt | b71636e | 2007-07-15 23:38:15 -0700 | [diff] [blame] | 2657 | /* Preserve the largest node */ |
| 2658 | if (largest < total_pages) { |
| 2659 | largest = total_pages; |
| 2660 | prefer = nid; |
| 2661 | } |
| 2662 | |
| 2663 | /* Interleave this node? */ |
| 2664 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) |
| 2665 | node_set(nid, interleave_nodes); |
| 2666 | } |
| 2667 | |
| 2668 | /* All too small, use the largest */ |
| 2669 | if (unlikely(nodes_empty(interleave_nodes))) |
| 2670 | node_set(prefer, interleave_nodes); |
| 2671 | |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 2672 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2673 | printk("numa_policy_init: interleaving failed\n"); |
Mel Gorman | 1a687c2 | 2012-11-22 11:16:36 +0000 | [diff] [blame] | 2674 | |
| 2675 | check_numabalancing_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2676 | } |
| 2677 | |
Christoph Lameter | 8bccd85 | 2005-10-29 18:16:59 -0700 | [diff] [blame] | 2678 | /* Reset policy of current process to default */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2679 | void numa_default_policy(void) |
| 2680 | { |
David Rientjes | 028fec4 | 2008-04-28 02:12:25 -0700 | [diff] [blame] | 2681 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2682 | } |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 2683 | |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 2684 | /* |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2685 | * Parse and format mempolicy from/to strings |
| 2686 | */ |
| 2687 | |
| 2688 | /* |
Hugh Dickins | f2a07f4 | 2013-01-02 02:01:33 -0800 | [diff] [blame] | 2689 | * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag. |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2690 | */ |
Lee Schermerhorn | 345ace9 | 2010-05-24 14:32:04 -0700 | [diff] [blame] | 2691 | static const char * const policy_modes[] = |
| 2692 | { |
| 2693 | [MPOL_DEFAULT] = "default", |
| 2694 | [MPOL_PREFERRED] = "prefer", |
| 2695 | [MPOL_BIND] = "bind", |
| 2696 | [MPOL_INTERLEAVE] = "interleave", |
Lee Schermerhorn | d3a7103 | 2012-10-25 14:16:29 +0200 | [diff] [blame] | 2697 | [MPOL_LOCAL] = "local", |
Lee Schermerhorn | 345ace9 | 2010-05-24 14:32:04 -0700 | [diff] [blame] | 2698 | }; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2699 | |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2700 | |
| 2701 | #ifdef CONFIG_TMPFS |
| 2702 | /** |
Hugh Dickins | f2a07f4 | 2013-01-02 02:01:33 -0800 | [diff] [blame] | 2703 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2704 | * @str: string containing mempolicy to parse |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2705 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2706 | * |
| 2707 | * Format of input: |
| 2708 | * <mode>[=<flags>][:<nodelist>] |
| 2709 | * |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2710 | * On success, returns 0, else 1 |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2711 | */ |
Hugh Dickins | a7a88b2 | 2013-01-02 02:04:23 -0800 | [diff] [blame] | 2712 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2713 | { |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2714 | struct mempolicy *new = NULL; |
Lee Schermerhorn | b4652e8 | 2010-05-24 14:32:03 -0700 | [diff] [blame] | 2715 | unsigned short mode; |
Hugh Dickins | f2a07f4 | 2013-01-02 02:01:33 -0800 | [diff] [blame] | 2716 | unsigned short mode_flags; |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2717 | nodemask_t nodes; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2718 | char *nodelist = strchr(str, ':'); |
| 2719 | char *flags = strchr(str, '='); |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2720 | int err = 1; |
| 2721 | |
| 2722 | if (nodelist) { |
| 2723 | /* NUL-terminate mode or flags string */ |
| 2724 | *nodelist++ = '\0'; |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2725 | if (nodelist_parse(nodelist, nodes)) |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2726 | goto out; |
Lai Jiangshan | 01f13bd | 2012-12-12 13:51:33 -0800 | [diff] [blame] | 2727 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2728 | goto out; |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2729 | } else |
| 2730 | nodes_clear(nodes); |
| 2731 | |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2732 | if (flags) |
| 2733 | *flags++ = '\0'; /* terminate mode string */ |
| 2734 | |
Peter Zijlstra | 479e280 | 2012-10-25 14:16:28 +0200 | [diff] [blame] | 2735 | for (mode = 0; mode < MPOL_MAX; mode++) { |
Lee Schermerhorn | 345ace9 | 2010-05-24 14:32:04 -0700 | [diff] [blame] | 2736 | if (!strcmp(str, policy_modes[mode])) { |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2737 | break; |
| 2738 | } |
| 2739 | } |
Mel Gorman | a720094 | 2012-11-16 09:37:58 +0000 | [diff] [blame] | 2740 | if (mode >= MPOL_MAX) |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2741 | goto out; |
| 2742 | |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2743 | switch (mode) { |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2744 | case MPOL_PREFERRED: |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2745 | /* |
| 2746 | * Insist on a nodelist of one node only |
| 2747 | */ |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2748 | if (nodelist) { |
| 2749 | char *rest = nodelist; |
| 2750 | while (isdigit(*rest)) |
| 2751 | rest++; |
KOSAKI Motohiro | 926f2ae | 2010-03-23 13:35:32 -0700 | [diff] [blame] | 2752 | if (*rest) |
| 2753 | goto out; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2754 | } |
| 2755 | break; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2756 | case MPOL_INTERLEAVE: |
| 2757 | /* |
| 2758 | * Default to online nodes with memory if no nodelist |
| 2759 | */ |
| 2760 | if (!nodelist) |
Lai Jiangshan | 01f13bd | 2012-12-12 13:51:33 -0800 | [diff] [blame] | 2761 | nodes = node_states[N_MEMORY]; |
Lee Schermerhorn | 3f226aa | 2008-04-28 02:13:24 -0700 | [diff] [blame] | 2762 | break; |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2763 | case MPOL_LOCAL: |
Lee Schermerhorn | 3f226aa | 2008-04-28 02:13:24 -0700 | [diff] [blame] | 2764 | /* |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2765 | * Don't allow a nodelist; mpol_new() checks flags |
Lee Schermerhorn | 3f226aa | 2008-04-28 02:13:24 -0700 | [diff] [blame] | 2766 | */ |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2767 | if (nodelist) |
Lee Schermerhorn | 3f226aa | 2008-04-28 02:13:24 -0700 | [diff] [blame] | 2768 | goto out; |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2769 | mode = MPOL_PREFERRED; |
Lee Schermerhorn | 3f226aa | 2008-04-28 02:13:24 -0700 | [diff] [blame] | 2770 | break; |
Ravikiran G Thirumalai | 413b43d | 2010-03-23 13:35:28 -0700 | [diff] [blame] | 2771 | case MPOL_DEFAULT: |
| 2772 | /* |
| 2773 | * Insist on a empty nodelist |
| 2774 | */ |
| 2775 | if (!nodelist) |
| 2776 | err = 0; |
| 2777 | goto out; |
KOSAKI Motohiro | d69b2e6 | 2010-03-23 13:35:30 -0700 | [diff] [blame] | 2778 | case MPOL_BIND: |
| 2779 | /* |
| 2780 | * Insist on a nodelist |
| 2781 | */ |
| 2782 | if (!nodelist) |
| 2783 | goto out; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2784 | } |
| 2785 | |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2786 | mode_flags = 0; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2787 | if (flags) { |
| 2788 | /* |
| 2789 | * Currently, we only support two mutually exclusive |
| 2790 | * mode flags. |
| 2791 | */ |
| 2792 | if (!strcmp(flags, "static")) |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2793 | mode_flags |= MPOL_F_STATIC_NODES; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2794 | else if (!strcmp(flags, "relative")) |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2795 | mode_flags |= MPOL_F_RELATIVE_NODES; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2796 | else |
KOSAKI Motohiro | 926f2ae | 2010-03-23 13:35:32 -0700 | [diff] [blame] | 2797 | goto out; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2798 | } |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2799 | |
| 2800 | new = mpol_new(mode, mode_flags, &nodes); |
| 2801 | if (IS_ERR(new)) |
KOSAKI Motohiro | 926f2ae | 2010-03-23 13:35:32 -0700 | [diff] [blame] | 2802 | goto out; |
| 2803 | |
Hugh Dickins | f2a07f4 | 2013-01-02 02:01:33 -0800 | [diff] [blame] | 2804 | /* |
| 2805 | * Save nodes for mpol_to_str() to show the tmpfs mount options |
| 2806 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. |
| 2807 | */ |
| 2808 | if (mode != MPOL_PREFERRED) |
| 2809 | new->v.nodes = nodes; |
| 2810 | else if (nodelist) |
| 2811 | new->v.preferred_node = first_node(nodes); |
| 2812 | else |
| 2813 | new->flags |= MPOL_F_LOCAL; |
| 2814 | |
| 2815 | /* |
| 2816 | * Save nodes for contextualization: this will be used to "clone" |
| 2817 | * the mempolicy in a specific context [cpuset] at a later time. |
| 2818 | */ |
| 2819 | new->w.user_nodemask = nodes; |
| 2820 | |
KOSAKI Motohiro | 926f2ae | 2010-03-23 13:35:32 -0700 | [diff] [blame] | 2821 | err = 0; |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2822 | |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2823 | out: |
| 2824 | /* Restore string for error message */ |
| 2825 | if (nodelist) |
| 2826 | *--nodelist = ':'; |
| 2827 | if (flags) |
| 2828 | *--flags = '='; |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2829 | if (!err) |
| 2830 | *mpol = new; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2831 | return err; |
| 2832 | } |
| 2833 | #endif /* CONFIG_TMPFS */ |
| 2834 | |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2835 | /** |
| 2836 | * mpol_to_str - format a mempolicy structure for printing |
| 2837 | * @buffer: to contain formatted mempolicy string |
| 2838 | * @maxlen: length of @buffer |
| 2839 | * @pol: pointer to mempolicy to be formatted |
Lee Schermerhorn | 71fe804 | 2008-04-28 02:13:26 -0700 | [diff] [blame] | 2840 | * |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2841 | * Convert a mempolicy into a string. |
| 2842 | * Returns the number of characters in buffer (if positive) |
| 2843 | * or an error (negative) |
| 2844 | */ |
Hugh Dickins | a7a88b2 | 2013-01-02 02:04:23 -0800 | [diff] [blame] | 2845 | int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2846 | { |
| 2847 | char *p = buffer; |
| 2848 | int l; |
| 2849 | nodemask_t nodes; |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 2850 | unsigned short mode; |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 2851 | unsigned short flags = pol ? pol->flags : 0; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2852 | |
Lee Schermerhorn | 2291990 | 2008-04-28 02:13:22 -0700 | [diff] [blame] | 2853 | /* |
| 2854 | * Sanity check: room for longest mode, flag and some nodes |
| 2855 | */ |
| 2856 | VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16); |
| 2857 | |
Lee Schermerhorn | bea904d | 2008-04-28 02:13:18 -0700 | [diff] [blame] | 2858 | if (!pol || pol == &default_policy) |
| 2859 | mode = MPOL_DEFAULT; |
| 2860 | else |
| 2861 | mode = pol->mode; |
| 2862 | |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2863 | switch (mode) { |
| 2864 | case MPOL_DEFAULT: |
| 2865 | nodes_clear(nodes); |
| 2866 | break; |
| 2867 | |
| 2868 | case MPOL_PREFERRED: |
| 2869 | nodes_clear(nodes); |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 2870 | if (flags & MPOL_F_LOCAL) |
Hugh Dickins | f2a07f4 | 2013-01-02 02:01:33 -0800 | [diff] [blame] | 2871 | mode = MPOL_LOCAL; |
Lee Schermerhorn | 53f2556 | 2008-04-28 02:13:20 -0700 | [diff] [blame] | 2872 | else |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 2873 | node_set(pol->v.preferred_node, nodes); |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2874 | break; |
| 2875 | |
| 2876 | case MPOL_BIND: |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2877 | /* Fall through */ |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2878 | case MPOL_INTERLEAVE: |
Hugh Dickins | f2a07f4 | 2013-01-02 02:01:33 -0800 | [diff] [blame] | 2879 | nodes = pol->v.nodes; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2880 | break; |
| 2881 | |
| 2882 | default: |
Dave Jones | 80de7c3 | 2012-09-06 12:01:00 -0400 | [diff] [blame] | 2883 | return -EINVAL; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2884 | } |
| 2885 | |
Lee Schermerhorn | 345ace9 | 2010-05-24 14:32:04 -0700 | [diff] [blame] | 2886 | l = strlen(policy_modes[mode]); |
Lee Schermerhorn | 53f2556 | 2008-04-28 02:13:20 -0700 | [diff] [blame] | 2887 | if (buffer + maxlen < p + l + 1) |
| 2888 | return -ENOSPC; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2889 | |
Lee Schermerhorn | 345ace9 | 2010-05-24 14:32:04 -0700 | [diff] [blame] | 2890 | strcpy(p, policy_modes[mode]); |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2891 | p += l; |
| 2892 | |
Lee Schermerhorn | fc36b8d | 2008-04-28 02:13:21 -0700 | [diff] [blame] | 2893 | if (flags & MPOL_MODE_FLAGS) { |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 2894 | if (buffer + maxlen < p + 2) |
| 2895 | return -ENOSPC; |
| 2896 | *p++ = '='; |
| 2897 | |
Lee Schermerhorn | 2291990 | 2008-04-28 02:13:22 -0700 | [diff] [blame] | 2898 | /* |
| 2899 | * Currently, the only defined flags are mutually exclusive |
| 2900 | */ |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 2901 | if (flags & MPOL_F_STATIC_NODES) |
Lee Schermerhorn | 2291990 | 2008-04-28 02:13:22 -0700 | [diff] [blame] | 2902 | p += snprintf(p, buffer + maxlen - p, "static"); |
| 2903 | else if (flags & MPOL_F_RELATIVE_NODES) |
| 2904 | p += snprintf(p, buffer + maxlen - p, "relative"); |
David Rientjes | f5b087b | 2008-04-28 02:12:27 -0700 | [diff] [blame] | 2905 | } |
| 2906 | |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2907 | if (!nodes_empty(nodes)) { |
| 2908 | if (buffer + maxlen < p + 2) |
| 2909 | return -ENOSPC; |
Lee Schermerhorn | 095f1fc | 2008-04-28 02:13:23 -0700 | [diff] [blame] | 2910 | *p++ = ':'; |
Christoph Lameter | 1a75a6c | 2006-01-08 01:01:02 -0800 | [diff] [blame] | 2911 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
| 2912 | } |
| 2913 | return p - buffer; |
| 2914 | } |